Learning inherently interpretable policies is a central challenge in the path to developing autonomous agents that humans can trust. Linear policies can justify their decisions while interacting in a dynamic environment, but their reduced expressivity prevents them from solving hard tasks. Instead, we argue for the use of piecewise-linear policies. We carefully study to what extent they can retain the interpretable properties of linear policies while reaching competitive performance with neural baselines. In particular, we propose the HyperCombinator (HC), a piecewise-linear neural architecture expressing a policy with a controllably small number of sub-policies. Each sub-policy is linear with respect to interpretable features, shedding light on the decision process of the agent without requiring an additional explanation model. We evaluate HC policies in control and navigation experiments, visualize the improved interpretability of the agent and highlight its trade-off with performance. Moreover, we validate that the restricted model class that the HyperCombinator belongs to is compatible with the algorithmic constraints of various reinforcement learning algorithms.

Foundation models are powerful technologies: how they are released publicly directly shapes their societal impact. In this position paper, we focus on *open* foundation models, defined here as those with broadly available model weights (e.g., Llama 3, Stable Diffusion XL). We identify five distinctive properties (e.g., greater customizability, poor monitoring) that mediate their benefits and risks. Open foundation models present significant benefits, with some caveats, that span innovation, competition, the distribution of decision-making power, and transparency. To understand their risks of misuse, we design a risk assessment framework for analyzing their *marginal risk*. Across several misuse vectors (e.g., cyberattacks, bioweapons), we find that current research is insufficient to effectively characterize the marginal risk of open foundation models relative to pre-existing technologies. The framework helps explain why the marginal risk is low in some cases, clarifies disagreements about misuse risks by revealing that past work has focused on different subsets of the framework with different assumptions, and articulates a way forward for more constructive debate. Overall, our work helps support a more grounded assessment of the societal impact of open foundation models by outlining what research is needed to empirically validate their theoretical benefits and risks.

Abstract We introduce ART, a new corpus-level autoencoding approach for training dense retrieval models that does not require any labeled training data. Dense retrieval is a central challenge for open-domain tasks, such as Open QA, where state-of-the-art methods typically require large supervised datasets with custom hard-negative mining and denoising of positive examples. ART, in contrast, only requires access to unpaired inputs and outputs (e.g., questions and potential answer passages). It uses a new passage-retrieval autoencoding scheme, where (1) an input question is used to retrieve a set of evidence passages, and (2) the passages are then used to compute the probability of reconstructing the original question. Training for retrieval based on question reconstruction enables effective unsupervised learning of both passage and question encoders, which can be later incorporated into complete Open QA systems without any further finetuning. Extensive experiments demonstrate that ART obtains state-of-the-art results on multiple QA retrieval benchmarks with only generic initialization from a pre-trained language model, removing the need for labeled data and task-specific losses.1 Our code and model checkpoints are available at: https://github.com/DevSinghSachan/art.

We propose a simple and effective re-ranking method for improving passage retrieval in open question answering.The re-ranker re-scores retrieved passages with a zero-shot question generation model, which uses a pre-trained language model to compute the probability of the input question conditioned on a retrieved passage.This approach can be applied on top of any retrieval method (e.g.neural or keywordbased), does not require any domain-or taskspecific training (and therefore is expected to generalize better to data distribution shifts), and provides rich cross-attention between query and passage (i.e. it must explain every token in the question).When evaluated on a number of open-domain retrieval datasets, our re-ranker improves strong unsupervised retrieval models by 6%-18% absolute and strong supervised models by up to 12% in terms of top-20 passage retrieval accuracy.We also obtain new stateof-the-art results on full open-domain question answering by simply adding the new re-ranker to existing models with no further changes.1

Reinforcement learning (RL) agents need to be robust to variations in safety-critical environments. While system identification methods provide a way to infer the variation from online experience, they can fail in settings where fast identification is not possible. Another dominant approach is robust RL which produces a policy that can handle worst-case scenarios, but these methods are generally designed to achieve robustness to a single uncertainty set that must be specified at train time. Towards a more general solution, we formulate the multi-set robustness problem to learn a policy robust to different perturbation sets. We then design an algorithm that enjoys the benefits of both system identification and robust RL: it reduces uncertainty where possible given a few interactions, but can still act robustly with respect to the remaining uncertainty. On a diverse set of control tasks, our approach demonstrates improved worst-case performance on new environments compared to prior methods based on system identification and on robust RL alone.

We propose a general framework for policy representation for reinforcement learning tasks. This framework involves finding a low-dimensional embedding of the policy on a reproducing kernel Hilbert space (RKHS). The usage of RKHS based methods allows us to derive strong theoretical guarantees on the expected return of the reconstructed policy. Such guarantees are typically lacking in black-box models, but are very desirable in tasks requiring stability and convergence guarantees. We conduct several experiments on classic RL domains. The results confirm that the policies can be robustly represented in a low-dimensional space while the embedded policy incurs almost no decrease in returns.

Training an agent to solve control tasks directly from high-dimensional images with model-free reinforcement learning (RL) has proven difficult. A promising approach is to learn a latent representation together with the control policy. However, fitting a high-capacity encoder using a scarce reward signal is sample inefficient and leads to poor performance. Prior work has shown that auxiliary losses, such as image reconstruction, can aid efficient representation learning. However, incorporating reconstruction loss into an off-policy learning algorithm often leads to training instability. We explore the underlying reasons and identify variational autoencoders, used by previous investigations, as the cause of the divergence. Following these findings, we propose effective techniques to improve training stability. This results in a simple approach capable of matching state-of-the-art model-free and model-based algorithms on MuJoCo control tasks. Furthermore, our approach demonstrates robustness to observational noise, surpassing existing approaches in this setting. Code, results, and videos are anonymously available at https://sites.google.com/view/sac-ae/home.

Koustuv Sinha, Prasanna Parthasarathi, Joelle Pineau, Adina Williams. Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers). 2021.

Real-world knowledge graphs are often characterized by low-frequency relations-a challenge that has prompted an increasing interest in few-shot link prediction methods.These methods perform link prediction for a set of new relations, unseen during training, given only a few example facts of each relation at test time.In this work, we perform a systematic study on a spectrum of models derived by generalizing the current state of the art for few-shot link prediction, with the goal of probing the limits of learning in this fewshot setting.We find that a simple zero-shot baseline-which ignores any relation-specific information-achieves surprisingly strong performance.Moreover, experiments on carefully crafted synthetic datasets show that having only a few examples of a relation fundamentally limits models from using fine-grained structural information and only allows for exploiting the coarse-grained positional information of entities.Together, our findings challenge the implicit assumptions and inductive biases of prior work and highlight new directions for research in this area.

A possible explanation for the impressive performance of masked language model (MLM) pre-training is that such models have learned to represent the syntactic structures prevalent in classical NLP pipelines.In this paper, we propose a different explanation: MLMs succeed on downstream tasks mostly due to their ability to model higher-order word cooccurrence statistics.To demonstrate this, we pre-train MLMs on sentences with randomly shuffled word order, and we show that these models still achieve high accuracy after finetuning on many downstream tasks -including tasks specifically designed to be challenging for models that ignore word order.Our models also perform surprisingly well according to some parametric syntactic probes, indicating possible deficiencies in how we test representations for syntactic information.Overall, our results show that purely distributional information largely explains the success of pretraining, and they underscore the importance of curating challenging evaluation datasets that require deeper linguistic knowledge.

We consider the offline reinforcement learning (RL) setting where the agent aims to optimize the policy solely from the data without further environment interactions. In offline RL, the distributional shift becomes the primary source of difficulty, which arises from the deviation of the target policy being optimized from the behavior policy used for data collection. This typically causes overestimation of action values, which poses severe problems for model-free algorithms that use bootstrapping. To mitigate the problem, prior offline RL algorithms often used sophisticated techniques that encourage underestimation of action values, which introduces an additional set of hyperparameters that need to be tuned properly. In this paper, we present an offline RL algorithm that prevents overestimation in a more principled way. Our algorithm, OptiDICE, directly estimates the stationary distribution corrections of the optimal policy and does not rely on policy-gradients, unlike previous offline RL algorithms. Using an extensive set of benchmark datasets for offline RL, we show that OptiDICE performs competitively with the state-of-the-art methods.

https://drive.google.com/file/d/1lRV72XaKoxZjgQrLXBJhsM82x54_1Vc4/view?usp=sharing

We study the problem of Safe Policy Improvement (SPI) under constraints in the offline Reinforcement Learning (RL) setting. We consider the scenario where: (i) we have a dataset collected under a known baseline policy, (ii) multiple reward signals are received from the environment inducing as many objectives to optimize. We present an SPI formulation for this RL setting that takes into account the preferences of the algorithm’s user for handling the trade-offs for different reward signals while ensuring that the new policy performs at least as well as the baseline policy along each individual objective. We build on traditional SPI algorithms and propose a novel method based on Safe Policy Iteration with Baseline Bootstrapping (SPIBB, Laroche et al., 2019) that provides high probability guarantees on the performance of the agent in the true environment. We show the effectiveness of our method on a synthetic grid-world safety task as well as in a real-world critical care context to learn a policy for the administration of IV fluids and vasopressors to treat sepsis.

Neural models trained for next utterance generation in dialogue task learn to mimic the n-gram sequences in the training set with training objectives like negative log-likelihood (NLL) or cross-entropy.Such commonly used training objectives do not foster generating alternate responses to a context.But, the effects of minimizing an alternate training objective that fosters a model to generate alternate response and score it on semantic similarity has not been well studied.We hypothesize that a language generation model can improve on its diversity by learning to generate alternate text during training and minimizing a semantic loss as an auxiliary objective.We explore this idea on two different sized data sets on the task of next utterance generation in goal oriented dialogues.We make two observations (1) minimizing a semantic objective improved diversity in responses in the smaller data set (Frames) but only as-good-as minimizing the NLL in the larger data set (Mul-tiWoZ) (2) large language model embeddings can be more useful as a semantic loss objective than as initialization for token embeddings.

Predicting the next utterance in dialogue is contingent on encoding of users' input text to generate appropriate and relevant response in data-driven approaches.Although the semantic and syntactic quality of the language generated is evaluated, more often than not, the encoded representation of input is not evaluated.As the representation of the encoder is essential for predicting the appropriate response, evaluation of encoder representation is a challenging yet important problem.In this work, we showcase evaluating the text generated through human or automatic metrics is not sufficient to appropriately evaluate soundness of the language understanding of dialogue models and, to that end, propose a set of probe tasks to evaluate encoder representation of different language encoders commonly used in dialogue models.From experiments, we observe that some of the probe tasks are easier and some are harder for even sophisticated model architectures to learn.And, through experiments we observe that RNN based architectures have lower performance on automatic metrics on text generation than transformer model but perform better than the transformer model on the probe tasks indicating that RNNs might preserve task information better than the Transformers.

One of the challenges in machine learning research is to ensure that presented and published results are sound and reliable. Reproducibility, that is obtaining similar results as presented in a paper or talk, using the same code and data (when available), is a necessary step to verify the reliability of research findings. Reproducibility is also an important step to promote open and accessible research, thereby allowing the scientific community to quickly integrate new findings and convert ideas to practice. Reproducibility also promotes the use of robust experimental workflows, which potentially reduce unintentional errors. In 2019, the Neural Information Processing Systems (NeurIPS) conference, the premier international conference for research in machine learning, introduced a reproducibility program, designed to improve the standards across the community for how we conduct, communicate, and evaluate machine learning research. The program contained three components: a code submission policy, a community-wide reproducibility challenge, and the inclusion of the Machine Learning Reproducibility checklist as part of the paper submission process. In this paper, we describe each of these components, how it was deployed, as well as what we were able to learn from this initiative.

The ability to detect and track objects in the visual world is a crucial skill for any intelligent agent, as it is a necessary precursor to any object-level reasoning process. Moreover, it is important that agents learn to track objects without supervision (i.e. without access to annotated training videos) since this will allow agents to begin operating in new environments with minimal human assistance. The task of learning to discover and track objects in videos, which we call unsupervised object tracking, has grown in prominence in recent years; however, most architectures that address it still struggle to deal with large scenes containing many objects. In the current work, we propose an architecture that scales well to the large-scene, many-object setting by employing spatially invariant computations (convolutions and spatial attention) and representations (a spatially local object specification scheme). In a series of experiments, we demonstrate a number of attractive features of our architecture; most notably, that it outperforms competing methods at tracking objects in cluttered scenes with many objects, and that it can generalize well to videos that are larger and/or contain more objects than videos encountered during training.

Applications of machine learning in biomedical prediction tasks are often limited by datasets that are unrepresentative of the sampling population. In these situations, we can no longer rely only on the the training data to learn the relations between features and the prediction outcome. Our method proposes to learn an inductive bias that indicates the relevance of each feature to outcomes through literature mining in PubMed, a centralized source of biomedical documents. The inductive bias acts as a source of prior knowledge from experts, which we leverage by imposing an extra penalty for model weights that differ from this inductive bias. We empirically evaluate our method on a medical prediction task and highlight the importance of incorporating expert knowledge that can capture relations not present in the training data.

Evaluating the quality of a dialogue interaction between two agents is a difficult task, especially in open-domain chit-chat style dialogue.There have been recent efforts to develop automatic dialogue evaluation metrics, but most of them do not generalize to unseen datasets and/or need a human-generated reference response during inference, making it infeasible for online evaluation.Here, we propose an unreferenced automated evaluation metric that uses large pre-trained language models to extract latent representations of utterances, and leverages the temporal transitions that exist between them.We show that our model achieves higher correlation with human annotations in an online setting, while not requiring true responses for comparison during inference.

Generalization across environments is critical to the successful application of reinforcement learning (RL) algorithms to real-world challenges. In this work we propose a method for learning state abstractions which generalize to novel observation distributions in the multi-environment RL setting. We prove that for certain classes of environments, this approach outputs, with high probability, a state abstraction corresponding to the causal feature set with respect to the return. We give empirical evidence that analogous methods for the nonlinear setting can also attain improved generalization over single- and multi-task baselines. Lastly, we provide bounds on model generalization error in the multi-environment setting, in the process showing a connection between causal variable identification and the state abstraction framework for MDPs.

We study the link between generalization and interference in temporal-difference (TD) learning. Interference is defined as the inner product of two different gradients, representing their alignment; this quantity emerges as being of interest from a variety of observations about neural networks, parameter sharing and the dynamics of learning. We find that TD easily leads to low-interference, under-generalizing parameters, while the effect seems reversed in supervised learning. We hypothesize that the cause can be traced back to the interplay between the dynamics of interference and bootstrapping. This is supported empirically by several observations: the negative relationship between the generalization gap and interference in TD, the negative effect of bootstrapping on interference and the local coherence of targets, and the contrast between the propagation rate of information in TD(0) versus TD($\lambda$) and regression tasks such as Monte-Carlo policy evaluation. We hope that these new findings can guide the future discovery of better bootstrapping methods.

We introduce and study the problem of Online Continual Compression, where one attempts to simultaneously learn to compress and store a representative dataset from a non i.i.d data stream, while only observing each sample once. A naive application of auto-encoder in this setting encounters a major challenge: representations derived from earlier encoder states must be usable by later decoder states. We show how to use discrete auto-encoders to effectively address this challenge and introduce Adaptive Quantization Modules (AQM) to control variation in the compression ability of the module at any given stage of learning. This enables selecting an appropriate compression for incoming samples, while taking into account overall memory constraints and current progress of the learned compression. Unlike previous methods, our approach does not require any pretraining, even on challenging datasets. We show that using AQM to replace standard episodic memory in continual learning settings leads to significant gains on continual learning benchmarks with images, LiDAR, and reinforcement learning agents.

Although Reinforcement Learning (RL) algorithms have found tremendous success in simulated domains, they often cannot directly be applied to physical systems, especially in cases where there are hard constraints to satisfy (e.g. on safety or resources). In standard RL, the agent is incentivized to explore any behavior as long as it maximizes rewards, but in the real world, undesired behavior can damage either the system or the agent in a way that breaks the learning process itself. In this work, we model the problem of learning with constraints as a Constrained Markov Decision Process and provide a new on-policy formulation for solving it. A key contribution of our approach is to translate cumulative cost constraints into state-based constraints. Through this, we define a safe policy improvement method which maximizes returns while ensuring that the constraints are satisfied at every step. We provide theoretical guarantees under which the agent converges while ensuring safety over the course of training. We also highlight the computational advantages of this approach. The effectiveness of our approach is demonstrated on safe navigation tasks and in safety-constrained versions of MuJoCo environments, with deep neural networks.

When an agent has limited information on its environment, the suboptimality of an RL algorithm can be decomposed into the sum of two terms: a term related to an asymptotic bias (suboptimality with unlimited data) and a term due to overfitting (additional suboptimality due to limited data). In the context of reinforcement learning with partial observability, this paper provides an analysis of the tradeoff between these two error sources. In particular, our theoretical analysis formally characterizes how a smaller state representation increases the asymptotic bias while decreasing the risk of overfitting.

By virtue of their expressive power, neural networks (NNs) are well suited to fitting large, complex datasets, yet they are also known to produce similar predictions for points outside the training distribution. As such, they are, like humans, under the influence of the Black Swan theory: models tend to be extremely "surprised" by rare events, leading to potentially disastrous consequences, while justifying these same events in hindsight. To avoid this pitfall, we introduce DENN, an ensemble approach building a set of Diversely Extrapolated Neural Networks that fits the training data and is able to generalize more diversely when extrapolating to novel data points. This leads DENN to output highly uncertain predictions for unexpected inputs. We achieve this by adding a diversity term in the loss function used to train the model, computed at specific inputs. We first illustrate the usefulness of the method on a low-dimensional regression problem. Then, we show how the loss can be adapted to tackle anomaly detection during classification, as well as safe imitation learning problems.

Adversarial Imitation Learning alternates between learning a discriminator -- which tells apart expert's demonstrations from generated ones -- and a generator's policy to produce trajectories that can fool this discriminator. This alternated optimization is known to be delicate in practice since it compounds unstable adversarial training with brittle and sample-inefficient reinforcement learning. We propose to remove the burden of the policy optimization steps by leveraging a novel discriminator formulation. Specifically, our discriminator is explicitly conditioned on two policies: the one from the previous generator's iteration and a learnable policy. When optimized, this discriminator directly learns the optimal generator's policy. Consequently, our discriminator's update solves the generator's optimization problem for free: learning a policy that imitates the expert does not require an additional optimization loop. This formulation effectively cuts by half the implementation and computational burden of Adversarial Imitation Learning algorithms by removing the Reinforcement Learning phase altogether. We show on a variety of tasks that our simpler approach is competitive to prevalent Imitation Learning methods.

We present a new approach for efficient exploration which leverages a low-dimensional encoding of the environment learned with a combination of model-based and model-free objectives. Our approach uses intrinsic rewards that are based on the distance of nearest neighbors in the low dimensional representational space to gauge novelty. We then leverage these intrinsic rewards for sample-efficient exploration with planning routines in representational space for hard exploration tasks with sparse rewards. One key element of our approach is the use of information theoretic principles to shape our representations in a way so that our novelty reward goes beyond pixel similarity. We test our approach on a number of maze tasks, as well as a control problem and show that our exploration approach is more sample-efficient compared to strong baselines.

Trust Region Policy Optimization (TRPO) and Proximal Policy Optimization (PPO) are among the most successful policy gradient approaches in deep reinforcement learning (RL). While these methods achieve state-of-the-art performance across a wide range of challenging tasks, there is room for improvement in the stabilization of the policy learning and how the off-policy data are used. In this paper we revisit the theoretical foundations of these algorithms and propose a new algorithm which stabilizes the policy improvement through a proximity term that constrains the discounted state-action visitation distribution induced by consecutive policies to be close to one another. This proximity term, expressed in terms of the divergence between the visitation distributions, is learned in an off-policy and adversarial manner. We empirically show that our proposed method can have a beneficial effect on stability and improve final performance in benchmark high-dimensional control tasks.

Deep reinforcement learning has recently shown many impressive successes. However, one major obstacle towards applying such methods to real-world problems is their lack of data-efficiency. To this end, we propose the Bottleneck Simulator: a model-based reinforcement learning method which combines a learned, factorized transition model of the environment with rollout simulations to learn an effective policy from few examples. The learned transition model employs an abstract, discrete (bottleneck) state, which increases sample efficiency by reducing the number of model parameters and by exploiting structural properties of the environment. We provide a mathematical analysis of the Bottleneck Simulator in terms of fixed points of the learned policy, which reveals how performance is affected by four distinct sources of error: an error related to the abstract space structure, an error related to the transition model estimation variance, an error related to the transition model estimation bias, and an error related to the transition model class bias. Finally, we evaluate the Bottleneck Simulator on two natural language processing tasks: a text adventure game and a real-world, complex dialogue response selection task. On both tasks, the Bottleneck Simulator yields excellent performance beating competing approaches.

There are many reasons to expect an ability to reason in terms of objects to be a crucial skill for any generally intelligent agent. Indeed, recent machine learning literature is replete with examples of the benefits of object-like representations: generalization, transfer to new tasks, and interpretability, among others. However, in order to reason in terms of objects, agents need a way of discovering and detecting objects in the visual world - a task which we call unsupervised object detection. This task has received significantly less attention in the literature than its supervised counterpart, especially in the case of large images containing many objects. In the current work, we develop a neural network architecture that effectively addresses this large-image, many-object setting. In particular, we combine ideas from Attend, Infer, Repeat (AIR), which performs unsupervised object detection but does not scale well, with recent developments in supervised object detection. We replace AIR’s core recurrent network with a convolutional (and thus spatially invariant) network, and make use of an object-specification scheme that describes the location of objects with respect to local grid cells rather than the image as a whole. Through a series of experiments, we demonstrate a number of features of our architecture: that, unlike AIR, it is able to discover and detect objects in large, many-object scenes; that it has a significant ability to generalize to images that are larger and contain more objects than images encountered during training; and that it is able to discover and detect objects with enough accuracy to facilitate non-trivial downstream processing.

Generative Adversarial Networks (GANs) can successfully approximate a probability distribution and produce realistic samples. However, open questions such as sufficient convergence conditions and mode collapse still persist. In this paper, we build on existing work in the area by proposing a novel framework for training the generator against an ensemble of discriminator networks, which can be seen as a one-student/multiple-teachers setting. We formalize this problem within the full-information adversarial bandit framework, where we evaluate the capability of an algorithm to select mixtures of discriminators for providing the generator with feedback during learning. To this end, we propose a reward function which reflects the progress made by the generator and dynamically update the mixture weights allocated to each discriminator. We also draw connections between our algorithm and stochastic optimization methods and then show that existing approaches using multiple discriminators in literature can be recovered from our framework. We argue that less expressive discriminators are smoother and have a general coarse grained view of the modes map, which enforces the generator to cover a wide portion of the data distribution support. On the other hand, highly expressive discriminators ensure samples quality. Finally, experimental results show that our approach improves samples quality and diversity over existing baselines by effectively learning a curriculum. These results also support the claim that weaker discriminators have higher entropy improving modes coverage.

In the quest for efficient and robust reinforcement learning methods, both model-free and model-based approaches offer advantages. In this paper we propose a new way of explicitly bridging both approaches via a shared low-dimensional learned encoding of the environment, meant to capture summarizing abstractions. We show that the modularity brought by this approach leads to good generalization while being computationally efficient, with planning happening in a smaller latent state space. In addition, this approach recovers a sufficient low-dimensional representation of the environment, which opens up new strategies for interpretable AI, exploration and transfer learning.

In this paper, we study the problem of multitask metric learning (mtML). We first examine the generalization bound of the regularized mtML formulation based on the notion of algorithmic stability, proving the convergence rate of mtML and revealing the trade-off between the tasks. Moreover, we also establish the theoretical connection between the mtML, single-task learning and pooling-task learning approaches. In addition, we present a novel boosting-based mtML (mt-BML) algorithm, which scales well with the feature dimension of the data. Finally, we also devise an efficient second-order Riemannian retraction operator which is tailored specifically to our mt-BML algorithm. It produces a low-rank solution of mtML to reduce the model complexity, and may also improve generalization performances. Extensive evaluations on several benchmark data sets verify the effectiveness of our learning algorithm.

Koustuv Sinha, Shagun Sodhani, Jin Dong, Joelle Pineau, William L. Hamilton. Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP). 2019.

We propose a targeted communication architecture for multi-agent reinforcement learning, where agents learn both what messages to send and whom to address them to while performing cooperative tasks in partially-observable environments. This targeting behavior is learnt solely from downstream task-specific reward without any communication supervision. We additionally augment this with a multi-round communication approach where agents coordinate via multiple rounds of communication before taking actions in the environment. We evaluate our approach on a diverse set of cooperative multi-agent tasks, of varying difficulties, with varying number of agents, in a variety of environments ranging from 2D grid layouts of shapes and simulated traffic junctions to 3D indoor environments, and demonstrate the benefits of targeted and multi-round communication. Moreover, we show that the targeted communication strategies learned by agents are interpretable and intuitive. Finally, we show that our architecture can be easily extended to mixed and competitive environments, leading to improved performance and sample complexity over recent state-of-the-art approaches.

Building models capable of generating structured output is a key challenge for AI and robotics. While generative models have been explored on many types of data, little work has been done on synthesizing lidar scans, which play a key role in robot mapping and localization. In this work, we show that one can adapt deep generative models for this task by unravelling lidar scans into a 2D point map. Our approach can generate high quality samples, while simultaneously learning a meaningful latent representation of the data. We demonstrate significant improvements against state-of-the-art point cloud generation methods. Furthermore, we propose a novel data representation that augments the 2D signal with absolute positional information. We show that this helps robustness to noisy and imputed input; the learned model can recover the underlying lidar scan from seemingly uninformative data.

Multi-simulator training has contributed to the recent success of Deep Reinforcement Learning (Deep RL) by stabilizing learning and allowing for higher training throughputs. In this work, we propose Gossip-based Actor-Learner Architectures (GALA) where several actor-learners (such as A2C agents) are organized in a peer-to-peer communication topology, and exchange information through asynchronous gossip in order to take advantage of a large number of distributed simulators. We prove that GALA agents remain within an epsilon-ball of one-another during training when using loosely coupled asynchronous communication. By reducing the amount of synchronization between agents, GALA is more computationally efficient and scalable compared to A2C, its fully-synchronous counterpart. GALA also outperforms A2C, being more robust and sample efficient. We show that we can run several loosely coupled GALA agents in parallel on a single GPU and achieve significantly higher hardware utilization and frame-rates than vanilla A2C at comparable power draws.

Diplomacy is a seven-player non-stochastic, non-cooperative game, where agents acquire resources through a mix of teamwork and betrayal. Reliance on trust and coordination makes Diplomacy the first non-cooperative multi-agent benchmark for complex sequential social dilemmas in a rich environment. In this work, we focus on training an agent that learns to play the No Press version of Diplomacy where there is no dedicated communication channel between players. We present DipNet, a neural-network-based policy model for No Press Diplomacy. The model was trained on a new dataset of more than 150,000 human games. Our model is trained by supervised learning (SL) from expert trajectories, which is then used to initialize a reinforcement learning (RL) agent trained through self-play. Both the SL and the RL agent demonstrate state-of-the-art No Press performance by beating popular rule-based bots.

This paper provides an analysis of the tradeoff between asymptotic bias (suboptimality with unlimited data) and overfitting (additional suboptimality due to limited data) in the context of reinforcement learning with partial observability. Our theoretical analysis formally characterizes that while potentially increasing the asymptotic bias, a smaller state representation decreases the risk of overfitting. This analysis relies on expressing the quality of a state representation by bounding $L_1$ error terms of the associated belief states. Theoretical results are empirically illustrated when the state representation is a truncated history of observations, both on synthetic POMDPs and on a large-scale POMDP in the context of smartgrids, with real-world data. Finally, similarly to known results in the fully observable setting, we also briefly discuss and empirically illustrate how using function approximators and adapting the discount factor may enhance the tradeoff between asymptotic bias and overfitting in the partially observable context.

Reinforcement learning has shown promise in learning policies that can solve complex problems. However, manually specifying a good reward function can be difficult, especially for intricate tasks. Inverse reinforcement learning offers a useful paradigm to learn the underlying reward function directly from expert demonstrations. Yet in reality, the corpus of demonstrations may contain trajectories arising from a diverse set of underlying reward functions rather than a single one. Thus, in inverse reinforcement learning, it is useful to consider such a decomposition. The options framework in reinforcement learning is specifically designed to decompose policies in a similar light. We therefore extend the options framework and propose a method to simultaneously recover reward options in addition to policy options. We leverage adversarial methods to learn joint reward-policy options using only observed expert states. We show that this approach works well in both simple and complex continuous control tasks and shows significant performance increases in one-shot transfer learning.

In recent years, significant progress has been made in solving challenging problems across various domains using deep reinforcement learning (RL). Reproducing existing work and accurately judging the improvements offered by novel methods is vital to sustaining this progress. Unfortunately, reproducing results for state-of-the-art deep RL methods is seldom straightforward. In particular, non-determinism in standard benchmark environments, combined with variance intrinsic to the methods, can make reported results tough to interpret. Without significance metrics and tighter standardization of experimental reporting, it is difficult to determine whether improvements over the prior state-of-the-art are meaningful. In this paper, we investigate challenges posed by reproducibility, proper experimental techniques, and reporting procedures. We illustrate the variability in reported metrics and results when comparing against common baselines and suggest guidelines to make future results in deep RL more reproducible. We aim to spur discussion about how to ensure continued progress in the field by minimizing wasted effort stemming from results that are non-reproducible and easily misinterpreted.

The use of dialogue systems as a medium for human-machine interaction is an increasingly prevalent paradigm. A growing number of dialogue systems use conversation strategies that are learned from large datasets. There are well documented instances where interactions with these system have resulted in biased or even offensive conversations due to the data-driven training process. Here, we highlight potential ethical issues that arise in dialogue systems research, including: implicit biases in data-driven systems, the rise of adversarial examples, potential sources of privacy violations, safety concerns, special considerations for reinforcement learning systems, and reproducibility concerns. We also suggest areas stemming from these issues that deserve further investigation. Through this initial survey, we hope to spur research leading to robust, safe, and ethically sound dialogue systems.

The use of connectionist approaches in conversational agents has been progressing rapidly due to the availability of large corpora.However current generative dialogue models often lack coherence and are content poor.This work proposes an architecture to incorporate unstructured knowledge sources to enhance the next utterance prediction in chit-chat type of generative dialogue models.We focus on Sequence-to-Sequence (Seq2Seq) conversational agents trained with the Reddit News dataset, and consider incorporating external knowledge from Wikipedia summaries as well as from the NELL knowledge base.Our experiments show faster training time and improved perplexity when leveraging external knowledge.

Recurrent Neural Networks (RNNs) with attention mechanisms have obtained state-of-the-art results for many sequence processing tasks. Most of these models use a simple form of encoder with attention that looks over the entire sequence and assigns a weight to each token independently. We present a mechanism for focusing RNN encoders for sequence modelling tasks which allows them to attend to key parts of the input as needed. We formulate this using a multi-layer conditional hierarchical sequence encoder that reads in one token at a time and makes a discrete decision on whether the token is relevant to the context or question being asked. The discrete gating mechanism takes in the context embedding and the current hidden state as inputs and controls information flow into the layer above. We train it using policy gradient methods. We evaluate this method on several types of tasks with different attributes. First, we evaluate the method on synthetic tasks which allow us to evaluate the model for its generalization ability and probe the behavior of the gates in more controlled settings. We then evaluate this approach on large scale Question Answering tasks including the challenging MS MARCO and SearchQA tasks. Our models shows consistent improvements for both tasks over prior work and our baselines. It has also shown to generalize significantly better on synthetic tasks as compared to the baselines.

In the pursuit of increasingly intelligent learning systems, abstraction plays a vital role in enabling sophisticated decisions to be made in complex environments. The options framework provides formalism for such abstraction over sequences of decisions. However most models require that options be given a priori, presumably specified by hand, which is neither efficient, nor scalable. Indeed, it is preferable to learn options directly from interaction with the environment. Despite several efforts, this remains a difficult problem. In this work we develop a novel policy gradient method for the automatic learning of policies with options. This algorithm uses inference methods to simultaneously improve all of the options available to an agent, and thus can be employed in an off-policy manner, without observing option labels. The differentiable inference procedure employed yields options that can be easily interpreted. Empirical results confirm these attributes, and indicate that our algorithm has an improved sample efficiency relative to state-of-the-art in learning options end-to-end.

Several applications of Reinforcement Learning suffer from instability due to high variance. This is especially prevalent in high dimensional domains. Regularization is a commonly used technique in machine learning to reduce variance, at the cost of introducing some bias. Most existing regularization techniques focus on spatial (perceptual) regularization. Yet in reinforcement learning, due to the nature of the Bellman equation, there is an opportunity to also exploit temporal regularization based on smoothness in value estimates over trajectories. This paper explores a class of methods for temporal regularization. We formally characterize the bias induced by this technique using Markov chain concepts. We illustrate the various characteristics of temporal regularization via a sequence of simple discrete and continuous MDPs, and show that the technique provides improvement even in high-dimensional Atari games.

We consider the problem of streaming kernel regression, when the observations arrive sequentially and the goal is to recover the underlying mean function, assumed to belong to an RKHS. The variance of the noise is not assumed to be known. In this context, we tackle the problem of tuning the regularization parameter adaptively at each time step, while maintaining tight confidence bounds estimates on the value of the mean function at each point. To this end, we first generalize existing results for finite-dimensional linear regression with fixed regularization and known variance to the kernel setup with a regularization parameter allowed to be a measurable function of past observations. Then, using appropriate self-normalized inequalities we build upper and lower bound estimates for the variance, leading to Bernstein-like concentration bounds. The latter is used in order to define the adaptive regularization. The bounds resulting from our technique are valid uniformly over all observation points and all time steps, and are compared against the literature with numerical experiments. Finally, the potential of these tools is illustrated by an application to kernelized bandits, where we revisit the Kernel UCB and Kernel Thompson Sampling procedures, and show the benefits of the novel adaptive kernel tuning strategy.

Sequential data often possesses hierarchical structures with complex dependencies between sub-sequences, such as found between the utterances in a dialogue. To model these dependencies in a generative framework, we propose a neural network-based generative architecture, with stochastic latent variables that span a variable number of time steps. We apply the proposed model to the task of dialogue response generation and compare it with other recent neural-network architectures. We evaluate the model performance through a human evaluation study. The experiments demonstrate that our model improves upon recently proposed models and that the latent variables facilitate both the generation of meaningful, long and diverse responses and maintaining dialogue state.

Ryan Lowe, Michael Noseworthy, Iulian Vlad Serban, Nicolas Angelard-Gontier, Yoshua Bengio, Joelle Pineau. Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2017.

Advances in neural variational inference have facilitated the learning of powerful directed graphical models with continuous latent variables, such as variational autoencoders.The hope is that such models will learn to represent rich, multi-modal latent factors in real-world data, such as natural language text.However, current models often assume simplistic priors on the latent variables -such as the uni-modal Gaussian distributionwhich are incapable of representing complex latent factors efficiently.To overcome this restriction, we propose the simple, but highly flexible, piecewise constant distribution.This distribution has the capacity to represent an exponential number of modes of a latent target distribution, while remaining mathematically tractable.Our results demonstrate that incorporating this new latent distribution into different models yields substantial improvements in natural language processing tasks such as document modeling and natural language generation for dialogue.

We consider the problem of estimating multiple related functions computed by weighted automata~(WFA). We first present a natural notion of relatedness between WFAs by considering to which extent several WFAs can share a common underlying representation. We then introduce the model of vector-valued WFA which conveniently helps us formalize this notion of relatedness. Finally, we propose a spectral learning algorithm for vector-valued WFAs to tackle the multitask learning problem. By jointly learning multiple tasks in the form of a vector-valued WFA, our algorithm enforces the discovery of a representation space shared between tasks. The benefits of the proposed multitask approach are theoretically motivated and showcased through experiments on both synthetic and real world datasets.

In goal-driven dialogue systems, success is often defined based on a structured definition of the goal.This requires that the dialogue system be constrained to handle a specific class of goals and that there be a mechanism to measure success with respect to that goal.However, in many human-human dialogues the diversity of goals makes it infeasible to define success in such a way.To address this scenario, we consider the task of automatically predicting success in goal-driven human-human dialogues using only the information communicated between participants in the form of text.We build a dataset from stackoverflow.com which consists of exchanges between two users in the technical domain where groundtruth success labels are available.We then propose a turn-based hierarchical neural network model that can be used to predict success without requiring a structured goal definition.We show this model outperforms rule-based heuristics and other baselines as it is able to detect patterns over the course of a dialogue and capture notions such as gratitude.

We propose a software architecture designed to ease the implementation of dialogue systems.The Modular Architecture for Conversational Agents (MACA) uses a plug-n-play style that allows quick prototyping, thereby facilitating the development of new techniques and the reproduction of previous work.The architecture separates the domain of the conversation from the agent's dialogue strategy, and as such can be easily extended to multiple domains.MACA provides tools to host dialogue agents on Amazon Mechanical Turk (mTurk) for data collection and allows processing of other sources of training data.The current version of the framework already incorporates several domains and existing dialogue strategies from the recent literature.

The dual-hormone artificial pancreas is an emerging technology to treat type 1 diabetes (T1D). It consists of a glucose sensor, infusion pumps, and a dosing algorithm that directs hormonal delivery. Preclinical optimization of dosing algorithms using computer simulations has the potential to accelerate the pace of development for this technology. However, current simulation environments consider glucose regulation models that either do not include glucagon action submodels or include submodels that were proposed without comparison to other candidate models. We consider here nine candidate models of glucagon action featuring a number of possible characteristics: insulin-independent glucagon action, insulin/glucagon ratio effect on hepatic glucose production, insulin-dependent suppression of glucagon action, and the effect of rate of change of glucagon. To assess the models, we use measurements of plasma insulin, plasma glucagon, and endogenous glucose production collected from experiments involving eight subjects with T1D who receive four subcutaneous glucagon boluses. We estimate each model's parameters using a Bayesian approach, and the models are contrasted based on the deviance information criterion. The model achieving the best fit features insulin-dependent suppression of glucagon action and incorporates effects of both glucagon levels and its rate of change.

A construct that has been receiving attention recently in reinforcement learning is stochastic factorization (SF), a particular case of non-negative factorization (NMF) in which the matrices involved are stochastic. The idea is to use SF to approximate the transition matrices of a Markov decision process (MDP). This is useful for two reasons. First, learning the factors of the SF instead of the transition matrices can reduce significantly the number of parameters to be estimated. Second, it has been shown that SF can be used to reduce the number of operations needed to compute an MDP's value function. Recently, an algorithm called expectation-maximization SF (EMSF) has been proposed to compute a SF directly from transitions sampled from an MDP. In this paper we take a closer look at EMSF. First, by exploiting the assumptions underlying the algorithm, we show that it is possible to reduce it to simple multiplicative update rules similar to the ones that helped popularize NMF. Second, we analyze the optimization process underlying EMSF and find that it minimizes a modified version of the Kullback-Leibler divergence that is particularly well-suited for learning a SF from data sampled from an arbitrary distribution. Third, we build on this improved understanding of EMSF to draw an interesting connection with NMF and probabilistic latent semantic analysis. We also exploit the simplified update rules to introduce a new version of EMSF that generalizes and significantly improves its precursor. This new algorithm provides a practical mechanism to control the trade-off between memory usage and computing time, essentially freeing the space complexity of EMSF from its dependency on the number of sample transitions. The algorithm can also compute its approximation incrementally, which makes it possible to use it concomitantly with the collection of data. This feature makes the new version of EMSF particularly suitable for online reinforcement learning. Empirical results support the utility of the proposed algorithm.

We investigate the task of building open domain, conversational dialogue systems based on large dialogue corpora using generative models. Generative models produce system responses that are autonomously generated word-by-word, opening up the possibility for realistic, flexible interactions. In support of this goal, we extend the recently proposed hierarchical recurrent encoder-decoder neural network to the dialogue domain, and demonstrate that this model is competitive with state-of-the-art neural language models and back-off n-gram models. We investigate the limitations of this and similar approaches, and show how its performance can be improved by bootstrapping the learning from a larger question-answer pair corpus and from pretrained word embeddings.

We present a novel multitask learning framework called multitask generalized eigenvalue program (MTGEP), which jointly solves multiple related generalized eigenvalue problems (GEPs). This framework is quite general and can be applied to many eigenvalue problems in machine learning and pattern recognition, ranging from supervised learning to unsupervised learning, such as principal component analysis (PCA), Fisher discriminant analysis (FDA), common spatial pattern (CSP), and so on. The core assumption of our approach is that the leading eigenvectors of related GEPs lie in some subspace that can be approximated by a sparse linear combination of basis vectors. As a result, these GEPs can be jointly solved by a sparse coding approach. Empirical evaluation with both synthetic and benchmark real world datasets validates the efficacy and efficiency of the proposed techniques, especially for grouped multitask GEPs.

We investigate evaluation metrics for dialogue response generation systems where supervised labels, such as task completion, are not available.Recent works in response generation have adopted metrics from machine translation to compare a model's generated response to a single target response.We show that these metrics correlate very weakly with human judgements in the non-technical Twitter domain, and not at all in the technical Ubuntu domain.We provide quantitative and qualitative results highlighting specific weaknesses in existing metrics, and provide recommendations for future development of better automatic evaluation metrics for dialogue systems.

Robot systems deployed in real-world environments often need to interact with other dynamic objects, such as pedestrians, cars, bicycles or other vehicles. In such cases, it is useful to have a good predictive model of the object's motion to factor in when optimizing the robot's own behaviour. In this paper we consider motion models cast in the Predictive Linear Gaussian (PLG) model, and propose two learning approaches for this framework: one based on the method of moments and the other on a least-squares criteria. We evaluate the approaches on several synthetic datasets, and deploy the system on a wheelchair robot, to improve its ability to follow a walking companion.

An open challenge in constructing dialogue systems is developing methods for automatically learning dialogue strategies from large amounts of unlabelled data. Recent work has proposed Next-Utterance-Classification (NUC) as a surrogate task for building dialogue systems from text data. In this paper we investigate the performance of humans on this task to validate the relevance of NUC as a method of evaluation. Our results show three main findings: (1) humans are able to correctly classify responses at a rate much better than chance, thus confirming that the task is feasible, (2) human performance levels vary across task domains (we consider 3 datasets) and expertise levels (novice vs experts), thus showing that a range of performance is possible on this type of task, (3) automated dialogue systems built using state-of-the-art machine learning methods have similar performance to the human novices, but worse than the experts, thus confirming the utility of this class of tasks for driving further research in automated dialogue systems.

Kernel-based reinforcement learning (KBRL) stands out among approximate reinforcement learning algorithms for its strong theoretical guarantees. By casting the learning problem as a local kernel approximation, KBRL provides a way of computing a decision policy which converges to a unique solution and is statistically consistent. Unfortunately, the model constructed by KBRL grows with the number of sample transitions, resulting in a computational cost that precludes its application to large-scale or on-line domains. In this paper we introduce an algorithm that turns KBRL into a practical reinforcement learning tool. Kernel-based stochastic factorization (KBSF) builds on a simple idea: when a transition probability matrix is represented as the product of two stochastic matrices, one can swap the factors of the multiplication to obtain another transition matrix, potentially much smaller than the original, which retains some fundamental properties of its precursor. KBSF exploits such an insight to compress the information contained in KBRL's model into an approximator of fixed size. This makes it possible to build an approximation considering both the difficulty of the problem and the associated computational cost. KBSF's computational complexity is linear in the number of sample transitions, which is the best one can do without discarding data. Moreover, the algorithm's simple mechanics allow for a fully incremental implementation that makes the amount of memory used independent of the number of sample transitions. The result is a kernel-based reinforcement learning algorithm that can be applied to large-scale problems in both off-line and on-line regimes. We derive upper bounds for the distance between the value functions computed by KBRL and KBSF using the same data. We also prove that it is possible to control the magnitude of the variables appearing in our bounds, which means that, given enough computational resources, we can make KBSF's value function as close as desired to the value function that would be computed by KBRL using the same set of sample transitions. The potential of our algorithm is demonstrated in an extensive empirical study in which KBSF is applied to difficult tasks based on real-world data. Not only does KBSF solve problems that had never been solved before, but it also significantly outperforms other state-of-the-art reinforcement learning algorithms on the tasks studied.

While both cost-sensitive learning and online learning have been studied separately, these two issues have seldom been addressed simultaneously. Yet, there are many applications where both aspects are important. This paper investigates a class of algorithmic approaches suitable for online cost-sensitive learning, designed for such problems. The basic idea is to leverage existing methods for online ensemble algorithms, and combine these with batch mode methods for cost-sensitive bagging/boosting algorithms. Within this framework, we describe several theoretically sound online cost-sensitive bagging and online cost-sensitive boosting algorithms, and show that the convergence of the proposed algorithms is guaranteed under certain conditions. We then present extensive experimental results on benchmark datasets to compare the performance of the various proposed approaches.

Planning in large partially observable Markov decision processes (POMDPs) is challenging especially when a long planning horizon is required. A few recent algorithms successfully tackle this case but at the expense of a weaker information-gathering capacity. In this paper, we propose Information Gathering and Reward Exploitation of Subgoals (IGRES), a randomized POMDP planning algorithm that leverages information in the state space to automatically generate "macro-actions" to tackle tasks with long planning horizons, while locally exploring the belief space to allow effective information gathering. Experimental results show that IGRES is an effective multi-purpose POMDP solver, providing state-of-the-art performance for both long horizon planning tasks and information-gathering tasks on benchmark domains. Additional experiments with an ecological adaptive management problem indicate that IGRES is a promising tool for POMDP planning in real-world settings.

The related problems of transfer learning and multitask learning have attracted significant attention, generating a rich literature of models and algorithms. Yet most existing approaches are studied in an offline fashion, implicitly assuming that data from different domains are given as a batch. Such an assumption is not valid in many real-world applications where data samples arrive sequentially, and one wants a good learner even from few examples. The goal of our work is to provide sound extensions to existing transfer and multitask learning algorithms such that they can be used in an anytime setting. More specifically, we propose two novel online boosting algorithms, one for transfer learning and one for multitask learning, both designed to leverage the knowledge of instances in other domains. The experimental results show state-of-the-art empirical performance on standard benchmarks, and we present results of using our methods for effectively detecting new seizures in patients with epilepsy from very few previous samples.

Having accurate knowledge of the positions of people around a robot provides rich, objective and quantitative data that can be highly useful for a wide range of tasks, including autonomous person following. The primary objective of this research is to promote the development of robust, repeatable and transferable software for robots that can automatically detect, track and follow people in their environment. The work is strongly motivated by the need for such functionality onboard an intelligent power wheelchair robot designed to assist people with mobility impairments. In this paper we propose a new algorithm for robust detection, tracking and following from laser data. We show that the approach is effective in various environments, both indoor and outdoor, and on different robot platforms (the intelligent power wheelchair and a Clearpath Husky). The method has been implemented in the Robot Operating System (ROS) framework and will be publicly released as a ROS package. We also describe and will release several datasets designed to promote the standardized evaluation of similar algorithms.

Wheelchairs play an important role for people living with locomotor impairments. However, power wheelchair users frequently report both minor and major accidents. The goal of this paper is to advocate for the use of robotic technology, in particular sensor-based detection and automatic classification of activities, to track and characterize activities onboard smart wheelchairs. Experiments were conducted in a clinical setting, in which experienced wheelchair users were asked to conduct a set of typical wheelchair activities. This paper presents an end-to-end pipeline for accurately classifying these activities from accelerometer data using signal processing and machine learning methods. Our classifier achieved an overall accuracy of around 50% in a more than 25 classes classification problem, compared to less than 4% with a random classifier. We also explored the possibility of discovering hidden patterns of activities using unsupervised topic modeling methods. We demonstrated the power of the inferred patterns with two use cases, namely story telling and hazard discovery. Altogether, this work provides new tools for characterizing the usage of smart wheelchairs with typical users.

This paper introduces the Ubuntu Dialogue Corpus, a dataset containing almost 1 million multi-turn dialogues, with a total of over 7 million utterances and 100 million words.This provides a unique resource for research into building dialogue managers based on neural language models that can make use of large amounts of unlabeled data.The dataset has both the multi-turn property of conversations in the Dialog State Tracking Challenge datasets, and the unstructured nature of interactions from microblog services such as Twitter.We also describe two neural learning architectures suitable for analyzing this dataset, and provide benchmark performance on the task of selecting the best next response.

Probabilistic latent-variable models are a powerful tool for modelling structured data. However, traditional expectation-maximization methods of learning such models are both computationally expensive and prone to local-minima. In contrast to these traditional methods, recently developed learning algorithms based upon the method of moments are both computationally efficient and provide strong statistical guarantees. In this work, we provide a unified presentation and empirical comparison of three general moment-based methods in the context of modelling stochastic languages. By rephrasing these methods upon a common theoretical ground, introducing novel theoretical results where necessary, we provide a clear comparison, making explicit the statistical assumptions upon which each method relies. With this theoretical grounding, we then provide an in-depth empirical analysis of the methods on both real and synthetic data with the goal of elucidating performance trends and highlighting important implementation details.

When a transition probability matrix is represented as the product of two stochastic matrices, one can swap the factors of the multiplication to obtain another transition matrix that retains some fundamental characteristics of the original. Since the derived matrix can be much smaller than its precursor, this property can be exploited to create a compact version of a Markov decision process (MDP), and hence to reduce the computational cost of dynamic programming. Building on this idea, this paper presents an approximate policy iteration algorithm called policy iteration based on stochastic factorization, or PISF for short. In terms of computational complexity, PISF replaces standard policy iteration's cubic dependence on the size of the MDP with a function that grows only linearly with the number of states in the model. The proposed algorithm also enjoys nice theoretical properties: it always terminates after a finite number of iterations and returns a decision policy whose performance only depends on the quality of the stochastic factorization. In particular, if the approximation error in the factorization is sufficiently small, PISF computes the optimal value function of the MDP. The paper also discusses practical ways of factoring an MDP and illustrates the usefulness of the proposed algorithm with an application involving a large-scale decision problem of real economical interest.

Learning accurate models of agent behaviours is crucial for the purpose of controlling systems where the agents' and environment's dynamics are unknown. This is a challenging problem, but structural assumptions can be leveraged to tackle it effectively. In particular, many systems exhibit mixed observability, when observations of some system components are essentially perfect and noiseless, while observations of other components are imperfect, aliased or noisy. In this paper we present a new model learning framework, the mixed observability predictive state representation (MO-PSR), which extends the previously known predictive state representations to the case of mixed observability systems. We present a learning algorithm that is scalable to large amounts of data and to large mixed observability domains, and show theoretical analysis of the learning consistency and computational complexity. Empirical results demonstrate that our algorithm is capable of learning accurate models, at a larger scale than with the generic predictive state representation, by leveraging the mixed observability properties.

Efficiently learning accurate models of dynamical systems is of central importance for developing rational agents that can succeed in a wide range of challenging domains. The difficulty of this learning problem is particularly acute in settings with large observation spaces and partial observability. We present a new algorithm, called Compressed Predictive State Representation (CPSR), for learning models of high-dimensional partially observable uncontrolled dynamical systems from small sample sets. The algorithm, which extends previous work on Predictive State Representations, exploits a particular sparse structure present in many domains. This sparse structure is used to compress information during learning, allowing for an increase in both the efficiency and predictive power. The compression technique also relieves the burden of domain specific feature selection and allows for domains with extremely large discrete observation spaces to be efficiently modelled. We present empirical results showing that the algorithm is able to build accurate models more efficiently than its uncompressed counterparts, and provide theoretical results on the accuracy of the learned compressed model.

This paper addresses the problem of automatic generation of features for value function approximation in reinforcement learning. Bellman Error Basis Functions (BEBFs) have been shown to improve the error of policy evaluation with function approximation, with a convergence rate similar to that of value iteration. We propose a simple, fast and robust algorithm based on random projections, which generates BEBFs for sparse feature spaces. We provide a finite sample analysis of the proposed method, and prove that projections logarithmic in the dimension of the original space guarantee a contraction in the error. Empirical results demonstrate the strength of this method in domains in which choosing a good state representation is challenging.

We propose an approach to learning from demonstration (LfD) which leverages expert data, even if the expert examples are very few or inaccurate. We achieve this by integrating LfD in an approximate policy iteration algorithm. The key idea of our approach is that expert examples are used to generate linear constraints on the optimization, in a similar fashion to large-margin classification. We prove an upper bound on the true Bellman error of the approximation computed by the algorithm at each iteration. We show empirically that the algorithm outperforms both pure policy iteration, as well as DAgger (a state-of-art LfD algorithm) and supervised learning in a variety of scenarios, including when very few and/or imperfect demonstrations are available. Our experiments include simulations as well as a real robotic navigation task.

We present a novel framework for analyzing univariate time series data. At the heart of the approach is a versatile algorithm for measuring the similarity of two segments of time series called geometric template matching (GeTeM). First, we use GeTeM to compute a similarity measure for clustering and nearest-neighbor classification. Next, we present a semi-supervised learning algorithm that uses the similarity measure with hierarchical clustering in order to improve classification performance when unlabeled training data are available. Finally, we present a boosting framework called TDEBOOST, which uses an ensemble of GeTeM classifiers. TDEBOOST augments the traditional boosting approach with an additional step in which the features used as inputs to the classifier are adapted at each step to improve the training error. We empirically evaluate the proposed approaches on several datasets, such as accelerometer data collected from wearable sensors and ECG data.

Recent advances in the area of compressed sensing suggest that it is possible to reconstruct high-dimensional sparse signals from a small number of random projections. Domains in which the sparsity assumption is applicable also offer many interesting large-scale machine learning prediction tasks. It is therefore important to study the effect of random projections as a dimensionality reduction method under such sparsity assumptions. In this paper we develop the bias-variance analysis of a least-squares regression estimator in compressed spaces when random projections are applied on sparse input signals. Leveraging the sparsity assumption, we are able to work with arbitrary non i.i.d. sampling strategies and derive a worst-case bound on the entire space. Empirical results on synthetic and real-world datasets shows how the choice of the projection size affects the performance of regression on compressed spaces, and highlights a range of problems where the method is useful.

The ability to learn a policy for a sequential decision problem with continuous state space using on-line data is a long-standing challenge. This paper presents a new reinforcement-learning algorithm, called iKBSF, which extends the benefits of kernel-based learning to the on-line scenario. As a kernel-based method, the proposed algorithm is stable and has good convergence properties. However, unlike other similar algorithms,iKBSF's space complexity is independent of the number of sample transitions, and as a result it can process an arbitrary amount of data. We present theoretical results showing that iKBSF can approximate (to any level of accuracy) the value function that would be learned by an equivalent batch non-parametric kernel-based reinforcement learning approximator. In order to show the effectiveness of the proposed algorithm in practice, we apply iKBSF to the challenging three-pole balancing task, where the ability to process a large number of transitions is crucial for achieving a high success rate.

The personalization of treatment via genetic biomarkers and other risk categories has drawn increasing interest among clinical researchers and scientists. A major challenge here is to construct individualized treatment rules (ITR), which recommend the best treatment for each of the different categories of individuals. In general, ITRs can be constructed using data from clinical trials, however these are generally very costly to run. In order to reduce the cost of learning an ITR, we explore active learning techniques designed to carefully decide whom to recruit, and which treatment to assign, throughout the online conduct of the clinical trial. As an initial investigation, we focus on simple ITRs that utilize a small number of subpopulation categories to personalize treatment. To minimize the maximal uncertainty regarding the treatment effects for each subpopulation, we propose the use of a minimax bandit model and provide an active learning policy for solving it. We evaluate our active learning policy using simulated data and data modeled after a clinical trial involving treatments for depressed individuals. We contrast this policy with other plausible active learning policies. The techniques presented in the paper may be generalized to tackle problems of efficient exploration in other domains.

Spoken dialogue systems are gaining popularity with improvements in speech recognition technologies. Dialogue systems can be modeled effectively using POMDPs, achieving improvements in robustness. However, past research on POMDPs-based dialogue system assumes that the model parameters are known. This limitation can be addressed through model-based Bayesian reinforcement learning, which offers a rich framework for simultaneous learning and planning. However, due to the high complexity of the framework, a major challenge is to scale up these algorithms for complex dialogue systems. In this work, we show that by exploiting certain known components of the system, such as knowledge of symmetrical properties, and using an approximate online planning algorithm, we are able to apply Bayesian RL on a realistic spoken dialogue system domain.

Kernel-based reinforcement-learning (KBRL) is a method for learning a decision policy from a set of sample transitions which stands out for its strong theoretical guarantees. However, the size of the approximator grows with the number of transitions, which makes the approach impractical for large problems. In this paper we introduce a novel algorithm to improve the scalability of KBRL. We resort to a special decomposition of a transition matrix, called stochastic factorization, to fix the size of the approximator while at the same time incorporating all the information contained in the data. The resulting algorithm, kernel-based stochastic factorization (KBSF), is much faster but still converges to a unique solution. We derive a theoretical upper bound for the distance between the value functions computed by KBRL and KBSF. The effectiveness of our method is illustrated with computational experiments on four reinforcement-learning problems, including a difficult task in which the goal is to learn a neurostimulation policy to suppress the occurrence of seizures in epileptic rat brains. We empirically demonstrate that the proposed approach is able to compress the information contained in KBRL's model. Also, on the tasks studied, KBSF outperforms two of the most prominent reinforcement-learning algorithms, namely least-squares policy iteration and fitted Q-iteration.

Markovian processes have long been used to model stochastic environments. Reinforcement learning has emerged as a framework to solve sequential planning and decision-making problems in such environments. In recent years, attempts were made to apply methods from reinforcement learning to construct decision support systems for action selection in Markovian environments. Although conventional methods in reinforcement learning have proved to be useful in problems concerning sequential decision-making, they cannot be applied in their current form to decision support systems, such as those in medical domains, as they suggest policies that are often highly prescriptive and leave little room for the user's input. Without the ability to provide flexible guidelines, it is unlikely that these methods can gain ground with users of such systems. This paper introduces the new concept of non-deterministic policies to allow more flexibility in the user's decision-making process, while constraining decisions to remain near optimal solutions. We provide two algorithms to compute non-deterministic policies in discrete domains. We study the output and running time of these method on a set of synthetic and real-world problems. In an experiment with human subjects, we show that humans assisted by hints based on non-deterministic policies outperform both human-only and computer-only agents in a web navigation task.

We present a method for automated topic suggestion. Given a plain-text input document, our algorithm produces a ranking of novel topics that could enrich the input document in a meaningful way. It can thus be used to assist human authors, who often fail to identify important topics relevant to the context of the documents they are writing. Our approach marries two algorithms originally designed for linking documents to Wikipedia articles, proposed by Milne and Witten [15] and West et al. [22]. While neither of them can suggest novel topics by itself, their combination does have this capability. The key step towards finding missing topics consists in generalizing from a large background corpus using principal component analysis. In a quantitative evaluation we conclude that our method achieves the precision of human editors when input documents are Wikipedia articles, and we complement this result with a qualitative analysis showing that the approach also works well on other types of input documents.

The problem of simultaneous localization and mapping (SLAM) is one of the most studied in the robotics literature. Most existing approaches, however, focus on scenarios where localization and mapping are the only tasks on the robot's agenda. In many real-world scenarios, a robot may be called on to perform other tasks simultaneously, in addition to localization and mapping. These can include target-following (or avoidance), search-and-rescue, point-to-point navigation, refueling, and so on. This paper proposes a framework that balances localization, mapping, and other planning objectives, thus allowing robots to solve sequential decision tasks under map and pose uncertainty. Our approach combines a SLAM algorithm with an online POMDP approach to solve diverse navigation tasks, without prior training, in an unknown environment.

Partially Observable Markov Decision Processes (POMDPs) offer a powerful mathematical framework for making optimal action choices in noisy and/or uncertain environments, in particular, allowing us to merge localization and decision-making for mobile robots. While advancements in POMDP techniques have allowed the use of much larger models, POMDPs for robot navigation are still limited by large state space requirements for even small maps. In this work, we propose a method to automatically generate a POMDP representation of an environment. By using variable resolution decomposition techniques, we can take advantage of characteristics of the environment to minimize the number of states required, while maintaining the level of detail required to find a robust and efficient policy. This is accomplished by automatically adjusting the level of detail required for planning at a given region, with few states representing large open areas, and many smaller states near objects. We validate this algorithm in POMDP simulations, a robot simulator as well as an autonomous robot.

This paper introduces the first set of PAC-Bayesian bounds for the batch reinforcement learning problem in finite state spaces. These bounds hold regardless of the correctness of the prior distribution. We demonstrate how such bounds can be used for model-selection in control problems where prior information is available either on the dynamics of the environment, or on the value of actions. Our empirical results confirm that PAC-Bayesian model-selection is able to leverage prior distributions when they are informative and, unlike standard Bayesian RL approaches, ignores them when they are misleading.

Wikipedia is the largest monolithic repository of human knowledge. In addition to its sheer size, it represents a new encyclopedic paradigm by interconnecting articles through hyperlinks. However, since these links are created by human authors, links one would expect to see are often missing. The goal of this work is to detect such gaps automatically. In this paper, we propose a novel method for augmenting the structure of hyperlinked document collections such as Wikipedia. It does not require the extraction of any manually defined features from the article to be augmented. Instead, it is based on principal component analysis, a well-founded mathematical generalization technique, and predicts new links purely based on the statistical structure of the graph formed by the existing links. Our method does not rely on the textual content of articles; we are exploiting only hyperlinks. A user evaluation of our technique shows that it improves the quality of top link suggestions over the state of the art and that the best predicted links are significantly more valuable than the 'average' link already present in Wikipedia. Beyond link prediction, our algorithm can potentially be used to point out topics an article misses to cover and to cluster articles semantically.

Personal robots are becoming increasingly prevalent, which raises a number of interesting issues regarding the design and customization of interfaces to such platforms. The particular problem addressed by this paper is the use of learning methods to improve the quality and effectiveness of human-machine interaction onboard a robotic wheelchair. In support of this, we present a method for learning and adapting probabilistic models with the aid of a human operator. We use a Bayesian reinforcement learning framework, that allows us to mix learning and execution, as well as take advantage of prior information about the world. We address the problems of learning, handling a partially observable environment, and limiting the number of action requests. We demonstrate empirical feasibility of our approach on an interface for an autonomous wheelchair.

Interesting real-world datasets often exhibit nonlinear, noisy, continuous-valued states that are unexplorable, are poorly described by first principles, and are only partially observable. If partial observability can be overcome, these constraints suggest the use of model-based reinforcement learning. We experiment with manifold embeddings as the reconstructed observable state-space of an off-line, model-based reinforcement learning approach to control. We demonstrate the embedding of a system changes as a result of learning and that the best performing embeddings well-represent the dynamics of both the uncontrolled and adaptively controlled system. We apply this approach in simulation to learn a neurostimulation policy that is more efficient in treating epilepsy than conventional policies. We then demonstrate the learned policy completely suppressing seizures in real-world neurostimulation experiments on actual animal brain slices.

This paper presents a new methodology for automatically learning an optimal neurostimulation strategy for the treatment of epilepsy. The technical challenge is to automatically modulate neurostimulation parameters, as a function of the observed EEG signal, so as to minimize the frequency and duration of seizures. The methodology leverages recent techniques from the machine learning literature, in particular the reinforcement learning paradigm, to formalize this optimization problem. We present an algorithm which is able to automatically learn an adaptive neurostimulation strategy directly from labeled training data acquired from animal brain tissues. Our results suggest that this methodology can be used to automatically find a stimulation strategy which effectively reduces the incidence of seizures, while also minimizing the amount of stimulation applied. This work highlights the crucial role that modern machine learning techniques can play in the optimization of treatment strategies for patients with chronic disorders such as epilepsy.

We consider the problem of optimal control in continuous and partially observable environments when the parameters of the model are not known exactly. Partially observable Markov decision processes (POMDPs) provide a rich mathematical model to handle such environments but require a known model to be solved by most approaches. This is a limitation in practice as the exact model parameters are often difficult to specify exactly. We adopt a Bayesian approach where a posterior distribution over the model parameters is maintained and updated through experience with the environment. We propose a particle filter algorithm to maintain the posterior distribution and an online planning algorithm, based on trajectory sampling, to plan the best action to perform under the current posterior. The resulting approach selects control actions which optimally trade-off between 1) exploring the environment to learn the model, 2) identifying the system's state, and 3) exploiting its knowledge in order to maximize long-term rewards. Our preliminary results on a simulated robot navigation problem show that our approach is able to learn good models of the sensors and actuators, and performs as well as if it had the true model.

Markov Decision Processes (MDPs) have been extensively studied and used in the context of planning and decision-making, and many methods exist to find the optimal policy for problems modelled as MDPs. Although finding the optimal policy is sufficient in many domains, in certain applications such as decision support systems where the policy is executed by a human (rather than a machine), finding all possible near-optimal policies might be useful as it provides more flexibility to the person executing the policy. In this paper we introduce the new concept of non-deterministic MDP policies, and address the question of finding near-optimal non-deterministic policies. We propose two solutions to this problem, one based on a Mixed Integer Program and the other one based on a search algorithm. We include experimental results obtained from applying this framework to optimize treatment choices in the context of a medical decision support system.

Partially Observable Markov Decision Processes (POMDPs) provide a rich framework for sequential decision-making under uncertainty in stochastic domains. However, solving a POMDP is often intractable except for small problems due to their complexity. Here, we focus on online approaches that alleviate the computational complexity by computing good local policies at each decision step during the execution. Online algorithms generally consist of a lookahead search to find the best action to execute at each time step in an environment. Our objectives here are to survey the various existing online POMDP methods, analyze their properties and discuss their advantages and disadvantages; and to thoroughly evaluate these online approaches in different environments under various metrics (return, error bound reduction, lower bound improvement). Our experimental results indicate that state-of-the-art online heuristic search methods can handle large POMDP domains efficiently.

While the partially observable Markov decision process (POMDP) provides a formal framework for the problem of robot control under uncertainty, it typically assumes a known and stationary model of the environment. In this paper, we study the problem of finding an optimal policy for controlling a robot in a partially observable domain, where the model is not perfectly known, and may change over time. We present an algorithm called MEDUSA which incrementally learns a POMDP model using queries, while still optimizing a reward function. We demonstrate effectiveness of the approach for a simple scenario, where a robot seeking a person has minimal a priori knowledge of its own sensor model, as well as where the person is located.

Bayesian Reinforcement Learning has generated substantial interest recently, as it provides an elegant solution to the exploration-exploitation trade-off in reinforce- ment learning. However most investigations of Bayesian reinforcement learning to date focus on the standard Markov Decision Processes (MDPs). Our goal is to extend these ideas to the more general Partially Observable MDP (POMDP) framework, where the state is a hidden variable. To address this problem, we in- troduce a new mathematical model, the Bayes-Adaptive POMDP. This new model allows us to (1) improve knowledge of the POMDP domain through interaction with the environment, and (2) plan optimal sequences of actions which can trade- off between improving the model, identifying the state, and gathering reward. We show how the model can be finitely approximated while preserving the value func- tion. We describe approximations for belief tracking and planning in this model. Empirical results on two domains show that the model estimate and agent’s return improve over time, as the agent learns better model estimates.

Planning in partially observable environments remains a challenging problem, despite significant recent advances in offline approximation techniques. A few online methods have also been proposed recently, and proven to be remarkably scalable, but without the theoretical guarantees of their offline counterparts. Thus it seems natural to try to unify offline and online techniques, preserving the theoretical properties of the former, and exploiting the scalability of the latter. In this paper, we provide theoretical guarantees on an anytime algorithm for POMDPs which aims to reduce the error made by approximate offline value iteration algorithms through the use of an efficient online searching procedure. The algorithm uses search heuristics based on an error analysis of lookahead search, to guide the online search towards reachable beliefs with the most potential to reduce error. We provide a general theorem showing that these search heuristics are admissible, and lead to complete and epsilon-optimal algorithms. This is, to the best of our knowledge, the strongest theoretical result available for online POMDP solution methods. We also provide empirical evidence showing that our approach is also practical, and can find (provably) near-optimal solutions in reasonable time.

The Partially Observable Markov Decision Process has long been recognized as a rich framework for real-world planning and control problems, especially in robotics. However exact solutions in this framework are typically computationally intractable for all but the smallest problems. A well-known technique for speeding up POMDP solving involves performing value backups at specific belief points, rather than over the entire belief simplex. The efficiency of this approach, however, depends greatly on the selection of points. This paper presents a set of novel techniques for selecting informative belief points which work well in practice. The point selection procedure is combined with point-based value backups to form an effective anytime POMDP algorithm called Point-Based Value Iteration (PBVI). The first aim of this paper is to introduce this algorithm and present a theoretical analysis justifying the choice of belief selection technique. The second aim of this paper is to provide a thorough empirical comparison between PBVI and other state-of-the-art POMDP methods, in particular the Perseus algorithm, in an effort to highlight their similarities and differences. Evaluation is performed using both standard POMDP domains and realistic robotic tasks.

Recent developments in grid-based and point-based approximation algo- rithms for POMDPs have greatly improved the tractability of POMDP planning. These approaches operate on sets of belief points by individ- ually learning a value function for each point. In reality, belief points exist in a highly-structured metric simplex, but current POMDP algo- rithms do not exploit this property. This paper presents a new metric-tree algorithm which can be used in the context of POMDP planning to sort belief points spatially, and then perform fast value function updates over groups of points. We present results showing that this approach can re- duce computation in point-based POMDP algorithms for a wide range of problems.

% '& ( *) + , -) .# / "0 1 2 3 4 5# 6 7 #

Dialog management is a critical component of an effective spoken language application. It is also one of the most difficult and time consuming to engineer. This paper examines the application of reinforcement learning and Markov decision processes (MDPs) to the problem of learning the dialog strategies. It extends work done at AT&T in two directions. First it examines the ability of RL to learn optimal strategies in the presence of speech recognition errors. Second, it describes a technique for reducing the amount of data required to train these models. This is significant as the difficulty of training MDP-based dialog managers is a serious roadblock to deploying them in realistic applications.