The performance of embodied agents has been shown to improve by increasing model parameters, dataset size, and compute. This has been demonstrated in domains from robotics to video games, when generative learning objectives on offline datasets (pre-training) are used to model an agent’s behavior (imitation learning) or their environment (world modeling). This paper characterizes the role of scale in these tasks more precisely. Going beyond the simple intuition that ‘bigger is better’, we show that the same types of power laws found in language modeling also arise in world modeling and imitation learning (e.g. between loss and optimal model size). However, the coefficients of these laws are heavily influenced by the tokenizer, task & architecture – this has important implications on the optimal sizing of models and data.

We propose Convex Constraint Learning for Reinforcement Learning (CoCoRL), a novel approach for inferring shared constraints in a Constrained Markov Decision Process (CMDP) from a set of safe demonstrations with possibly different reward functions. While previous work is limited to demonstrations with known rewards or fully known environment dynamics, CoCoRL can learn constraints from demonstrations with different unknown rewards without knowledge of the environment dynamics. CoCoRL constructs a convex safe set based on demonstrations, which provably guarantees safety even for potentially sub-optimal (but safe) demonstrations. For near-optimal demonstrations, CoCoRL converges to the true safe set with no policy regret. We evaluate CoCoRL in gridworld environments and a driving simulation with multiple constraints. CoCoRL learns constraints that lead to safe driving behavior. Importantly, we can safely transfer the learned constraints to different tasks and environments. In contrast, alternative methods based on Inverse Reinforcement Learning (IRL) often exhibit poor performance and learn unsafe policies.

We aim to understand how people assess human likeness in navigation produced by people and artificially intelligent (AI) agents in a video game. To this end, we propose a novel AI agent with the goal of generating more human-like behavior. We collect hundreds of crowd-sourced assessments comparing the human-likeness of navigation behavior generated by our agent and baseline AI agents with human-generated behavior. Our proposed agent passes a Turing Test, while the baseline agents do not. By passing a Turing Test, we mean that human judges could not quantitatively distinguish between videos of a person and an AI agent navigating. To understand what people believe constitutes human-like navigation, we extensively analyze the justifications of these assessments. This work provides insights into the characteristics that people consider human-like in the context of goal-directed video game navigation, which is a key step for further improving human interactions with AI agents.

Sample efficiency is crucial for imitation learning methods to be applicable in real-world applications. Many studies improve sample efficiency by extending adversarial imitation to be off-policy regardless of the fact that these off-policy extensions could either change the original objective or involve complicated optimization. We revisit the foundation of adversarial imitation and propose an off-policy sample efficient approach that requires no adversarial training or min-max optimization. Our formulation capitalizes on two key insights: (1) the similarity between the Bellman equation and the stationary state-action distribution equation allows us to derive a novel temporal difference (TD) learning approach; and (2) the use of a deterministic policy simplifies the TD learning. Combined, these insights yield a practical algorithm, Deterministic and Discriminative Imitation (D2-Imitation), which oper- ates by first partitioning samples into two replay buffers and then learning a deterministic policy via off-policy reinforcement learning. Our empirical results show that D2-Imitation is effective in achieving good sample efficiency, outperforming several off-policy extension approaches of adversarial imitation on many control tasks.

We study sequential decision-making with known rewards and unknown constraints, motivated by situations where the constraints represent expensive-to-evaluate human preferences, such as safe and comfortable driving behavior. We formalize the challenge of interactively learning about these constraints as a novel linear bandit problem which we call constrained linear best-arm identification. To solve this problem, we propose the Adaptive Constraint Learning (ACOL) algorithm. We provide an instance-dependent lower bound for constrained linear best-arm identification and show that ACOL’s sample complexity matches the lower bound in the worst-case. In the average case, ACOL’s sample complexity bound is still significantly tighter than bounds of simpler approaches. In synthetic experiments, ACOL performs on par with an oracle solution and outperforms a range of baselines. As an application, we consider learning constraints to represent human preferences in a driving simulation. ACOL is significantly more sample efficient than alternatives for this application. Further, we find that learning preferences as constraints is more robust to changes in the driving scenario than encoding the preferences directly in the reward function.

Randomly masking and predicting word tokens has been a successful approach in pre-training language models for a variety of downstream tasks. In this work, we observe that the same idea also applies naturally to sequential decision making, where many well-studied tasks like behavior cloning, offline RL, inverse dynamics, and waypoint conditioning correspond to different sequence maskings over a sequence of states, actions, and returns. We introduce the UniMASK framework, which provides a unified way to specify models which can be trained on many different sequential decision making tasks. We show that a single UniMASK model is often capable of carrying out many tasks with performance similar to or better than single-task models. Additionally, after fine-tuning, our UniMASK models consistently outperform comparable single-task models.

Recent years have seen a growth in user-centric applications that require effective knowledge transfer across tasks in the low-data regime. An example is personalization, where a pretrained system is adapted by learning on small amounts of labeled data belonging to a specific user. This setting requires high accuracy under low computational complexity, therefore the Pareto frontier of accuracy vs. adaptation cost plays a crucial role. In this paper we push this Pareto frontier in the few-shot image classification setting with a key contribution: a new adaptive block called Contextual Squeeze-and-Excitation (CaSE) that adjusts a pretrained neural network on a new task to significantly improve performance with a single forward pass of the user data (context). We use meta-trained CaSE blocks to conditionally adapt the body of a network and a fine-tuning routine to adapt a linear head, defining a method called UpperCaSE. UpperCaSE achieves a new state-of-the-art accuracy relative to meta-learners on the 26 datasets of VTAB+MD and on a challenging real-world personalization benchmark (ORBIT), narrowing the gap with leading fine-tuning methods with the benefit of orders of magnitude lower adaptation cost.

Artificial Intelligence (AI) for accessibility is a rapidly growing area, requiring datasets that are inclusive of the disabled users that assistive technology aims to serve. We offer insights from a multi-disciplinary project that constructed a dataset for teachable object recognition with people who are blind or low vision. Teachable object recognition enables users to teach a model objects that are of interest to them, e.g., their white cane or own sunglasses, by providing example images or videos of objects. In this paper, we make the following contributions: 1) a disability-first procedure to support blind and low vision data collectors to produce good quality data, using video rather than images; 2) a validation and evolution of this procedure through a series of data collection phases and 3) a set of questions to orient researchers involved in creating datasets toward reflecting on the needs of their participant community.

Object recognition has made great advances in the last decade, but predominately still relies on many high-quality training examples per object category. In contrast, learning new objects from only a few examples could enable many impactful applications from robotics to user personalization. Most few-shot learning research, however, has been driven by benchmark datasets that lack the high variation that these applications will face when deployed in the real-world. To close this gap, we present the ORBIT dataset and benchmark, grounded in the real-world application of teachable object recognizers for people who are blind/low-vision. The dataset contains 3,822 videos of 486 objects recorded by people who are blind/low-vision on their mobile phones. The benchmark reflects a realistic, highly challenging recognition problem, providing a rich playground to drive research in robustness to few-shot, high-variation conditions. We set the benchmark’s first state-of-the-art and show there is massive scope for further innovation, holding the potential to impact a broad range of real-world vision applications including tools for the blind/low-vision community. We release the dataset at https://doi.org/10.25383/city.14294597 and benchmark code at https://github.com/microsoft/ORBIT-Dataset.

A key challenge on the path to developing agents that learn complex human-like behavior is the need to quickly and accurately quantify human-likeness. While human assessments of such behavior can be highly accurate, speed and scalability are limited. We address these limitations through a novel automated Navigation Turing Test (ANTT) that learns to predict human judgments of human-likeness. We demonstrate the effectiveness of our automated NTT on a navigation task in a complex 3D environment. We investigate six classification models to shed light on the types of architectures best suited to this task, and validate them against data collected through a human NTT. Our best models achieve high accuracy when distinguishing true human and agent behavior. At the same time, we show that predicting finer-grained human assessment of agents’ progress towards human-like behavior remains unsolved. Our work takes an important step towards agents that more effectively learn complex human-like behavior.

Training autonomous agents able to generalize to multiple tasks is a key target of Deep Reinforcement Learning (DRL) research. In parallel to improving DRL algorithms themselves, Automatic Curriculum Learning (ACL) study how teacher algorithms can train DRL agents more efficiently by adapting task selection to their evolving abilities. While multiple standard benchmarks exist to compare DRL agents, there is currently no such thing for ACL algorithms. Thus, comparing existing approaches is difficult, as too many experimental parameters differ from paper to paper. In this work, we identify several key challenges faced by ACL algorithms. Based on these, we present TeachMyAgent (TA), a benchmark of current ACL algorithms leveraging procedural task generation. It includes 1) challenge-specific unit-tests using variants of a procedural Box2D bipedal walker environment, and 2) a new procedural Parkour environment combining most ACL challenges, making it ideal for global performance assessment. We then use TeachMyAgent to conduct a comparative study of representative existing approaches, showcasing the competitiveness of some ACL algorithms that do not use expert knowledge. We also show that the Parkour environment remains an open problem. We open-source our environments, all studied ACL algorithms (collected from open-source code or re-implemented), and DRL students in a Python package available at https://github.com/flowersteam/TeachMyAgent.

To rapidly learn a new task, it is often essential for agents to explore efficiently - especially when performance matters from the first timestep. One way to learn such behaviour is via meta-learning. Many existing methods however rely on dense rewards for meta-training, and can fail catastrophically if the rewards are sparse. Without a suitable reward signal, the need for exploration during meta-training is exacerbated. To address this, we propose HyperX, which uses novel reward bonuses for meta-training to explore in approximate hyper-state space (where hyper-states represent the environment state and the agent’s task belief). We show empirically that HyperX meta-learns better task-exploration and adapts more successfully to new tasks than existing methods.

Meta learning approaches to few-shot classification are computationally efficient at test time, requiring just a few optimization steps or single forward pass to learn a new task, but they remain highly memory-intensive to train. This limitation arises because a task's entire support set, which can contain up to 1000 images, must be processed before an optimization step can be taken. Harnessing the performance gains offered by large images thus requires either parallelizing the meta-learner across multiple GPUs, which may not be available, or trade-offs between task and image size when memory constraints apply. We improve on both options by proposing LITE, a general and memory efficient episodic training scheme that enables meta-training on large tasks composed of large images on a single GPU. We achieve this by observing that the gradients for a task can be decomposed into a sum of gradients over the task's training images. This enables us to perform a forward pass on a task's entire training set but realize significant memory savings by back-propagating only a random subset of these images which we show is an unbiased approximation of the full gradient. We use LITE to train meta-learners and demonstrate new state-of-the-art accuracy on the real-world ORBIT benchmark and 3 of the 4 parts of the challenging VTAB+MD benchmark relative to leading meta-learners. LITE also enables meta-learners to be competitive with transfer learning approaches but at a fraction of the test-time computational cost, thus serving as a counterpoint to the recent narrative that transfer learning is all you need for few-shot classification.

Language is an interface to the outside world. In order for embodied agents to use it, language must be grounded in other, sensorimotor modalities. While there is an extended literature studying how machines can learn grounded language, the topic of how to learn spatio-temporal linguistic concepts is still largely uncharted. To make progress in this direction, we here introduce a novel spatio-temporal language grounding task where the goal is to learn the meaning of spatio-temporal descriptions of behavioral traces of an embodied agent. This is achieved by training a truth function that predicts if a description matches a given history of observations. The descriptions involve time-extended predicates in past and present tense as well as spatio-temporal references to objects in the scene. To study the role of architectural biases in this task, we train several models including multimodal Transformer architectures; the latter implement different attention computations between words and objects across space and time. We test models on two classes of generalization: 1) generalization to new sentences, 2) generalization to grammar primitives. We observe that maintaining object identity in the attention computation of our Transformers is instrumental to achieving good performance on generalization overall, and that summarizing object traces in a single token has little influence on performance. We then discuss how this opens new perspectives for language-guided autonomous embodied agents.

High sample complexity remains a barrier to the application of reinforcement learning (RL), particularly in multi-agent systems. A large body of work has demonstrated that exploration mechanisms based on the principle of optimism under uncertainty can significantly improve the sample efficiency of RL in single agent tasks. This work seeks to understand the role of optimistic exploration in non-cooperative multi-agent settings. We will show that, in zero-sum games, optimistic exploration can cause the learner to waste time sampling parts of the state space that are irrelevant to strategic play, as they can only be reached through cooperation between both players. To address this issue, we introduce a formal notion of strategically efficient exploration in Markov games, and use this to develop two strategically efficient learning algorithms for finite Markov games. We demonstrate that these methods can be significantly more sample efficient than their optimistic counterparts.

Trading off exploration and exploitation in an unknown environment is key to maximising expected online return during learning. A Bayes-optimal policy, which does so optimally, conditions its actions not only on the environment state but also on the agent's uncertainty about the environment. Computing a Bayes-optimal policy is however intractable for all but the smallest tasks. In this paper, we introduce variational Bayes-Adaptive Deep RL (variBAD), a way to meta-learn approximately Bayes-optimal policies for complex tasks. VariBAD simultaneously meta-learns a variational auto-encoder to perform approximate inference, and a policy that incorporates task uncertainty directly during action selection by conditioning on both the environment state and the approximate belief. In two toy domains, we illustrate how variBAD performs structured online exploration as a function of task uncertainty. We further evaluate variBAD on MuJoCo tasks widely used in meta-RL and show that it achieves higher online return than existing methods. On the recently proposed Meta-World ML1 benchmark, variBAD achieves state of the art results by a large margin, fully solving two out of the three ML1 tasks for the first time.

Learning workable representations of dynamical systems is becoming an increasingly important problem in a number of application areas. By leveraging recent work connecting deep neural networks to systems of differential equations, we propose \emph{variational integrator networks}, a class of neural network architectures designed to preserve the geometric structure of physical systems. This class of network architectures facilitates accurate long-term prediction, interpretability, and data-efficient learning, while still remaining highly flexible and capable of modeling complex behavior. We demonstrate that they can accurately learn dynamical systems from both noisy observations in phase space and from image pixels within which the unknown dynamics are embedded.

Automatic Curriculum Learning (ACL) has become a cornerstone of recent successes in Deep Reinforcement Learning (DRL). These methods shape the learning trajectories of agents by challenging them with tasks adapted to their capacities. In recent years, they have been used to improve sample efficiency and asymptotic performance, to organize exploration, to encourage generalization or to solve sparse reward problems, among others. To do so, ACL mechanisms can act on many aspects of learning problems. They can optimize domain randomization for Sim2Real transfer, organize task presentations in multi-task robotic settings, order sequences of opponents in multi-agent scenarios, etc. The ambition of this work is dual: 1) to present a compact and accessible introduction to the Automatic Curriculum Learning literature and 2) to draw a bigger picture of the current state of the art in ACL to encourage the cross-breeding of existing concepts and the emergence of new ideas.

This paper presents a customisable 3D benchmark for assessing generalisability of reinforcement learning agents based on the 3D first-person game Doom and open source environment VizDoom. As a sample use-case we show that different domain randomisation techniques during training in a key-collection navigation task can help to improve agent performance on unseen evaluation maps.

AI agents within video games are often required to compete within an environment shared by many other agents. This problem can be tackled by multi-agent reinforcement learning (MARL). One solution to MARL is to learn a Nash Equilibrium Strategy (NES) that guarantees a known minimum payoff when playing against other rational agents. We focus on one approach for learning a NES, Win or Learn Fast (WoLF), WoLF has been shown to converge towards a NES in a variety of matrix-games and grid based games. Research into Deep MARL has focused on performance against opponent agents and with limited quantitative results regarding learning a NES. We present a systematic empirical investigation into the ability of Proximal Policy Optimisation (PPO) to learn a NES, showing instability in certain matrix games. We then present an extension, WoLF-PPO, that is able to learn a policy that is closer to the NES.

We propose CAVIA for meta-learning, a simple extension to MAML that is less prone to meta-overfitting, easier to parallelise, and more interpretable. CAVIA partitions the model parameters into two parts: context parameters that serve as additional input to the model and are adapted on individual tasks, and shared parameters that are meta-trained and shared across tasks. At test time, only the context parameters are updated, leading to a low-dimensional task representation. We show empirically that CAVIA outperforms MAML for regression, classification, and reinforcement learning. Our experiments also highlight weaknesses in current benchmarks, in that the amount of adaptation needed in some cases is small.

Actor-critic methods, a type of model-free Reinforcement Learning, have been successfully applied to challenging tasks in continuous control, often achieving state-of-the art performance. However, wide-scale adoption of these methods in real-world domains is made difficult by their poor sample efficiency. We address this problem both theoretically and empirically. On the theoretical side, we identify two phenomena preventing efficient exploration in existing state-of-the-art algorithms such as Soft Actor Critic. First, combining a greedy actor update with a pessimistic estimate of the critic leads to the avoidance of actions that the agent does not know about, a phenomenon we call pessimistic underexploration. Second, current algorithms are directionally uninformed, sampling actions with equal probability in opposite directions from the current mean. This is wasteful, since we typically need actions taken along certain directions much more than others. To address both of these phenomena, we introduce a new algorithm, Optimistic Actor Critic, which approximates a lower and upper confidence bound on the state-action value function. This allows us to apply the principle of optimism in the face of uncertainty to perform directed exploration using the upper bound while still using the lower bound to avoid overestimation. We evaluate OAC in several challenging continuous control tasks, achieving state-of the art sample efficiency.

The ability for policies to generalize to new environments is key to the broad application of RL agents. A promising approach to prevent an agent’s policy from overfitting to a limited set of training environments is to apply regularization techniques originally developed for supervised learning. However, there are stark differences between supervised learning and RL. We discuss those differences and propose modifications to existing regularization techniques in order to better adapt them to RL. In particular, we focus on regularization techniques relying on the injection of noise into the learned function, a family that includes some of the most widely used approaches such as Dropout and Batch Normalization. To adapt them to RL, we propose Selective Noise Injection (SNI), which maintains the regularizing effect the injected noise has, while mitigating the adverse effects it has on the gradient quality. Furthermore, we demonstrate that the Information Bottleneck (IB) is a particularly well suited regularization technique for RL as it is effective in the low-data regime encountered early on in training RL agents. Combining the IB with SNI, we significantly outperform current state of the art results, including on the recently proposed generalization benchmark Coinrun.

Posterior sampling for reinforcement learning (PSRL) is an effective method for balancing exploration and exploitation in reinforcement learning. Randomised value functions (RVF) can be viewed as a promising approach to scaling PSRL. However, we show that most contemporary algorithms combining RVF with neural network function approximation do not possess the properties which make PSRL effective, and provably fail in sparse reward problems. Moreover, we find that propagation of uncertainty, a property of PSRL previously thought important for exploration, does not preclude this failure. We use these insights to design Successor Uncertainties (SU), a cheap and easy to implement RVF algorithm that retains key properties of PSRL. SU is highly effective on hard tabular exploration benchmarks. Furthermore, on the Atari 2600 domain, it surpasses human performance on 38 of 49 games tested (achieving a median human normalised score of 2.09), and outperforms its closest RVF competitor, Bootstrapped DQN, on 36 of those.

The dueling bandits problem is an online learning framework where learning happens ``on-the-fly'' through preference feedback, i.e., from comparisons between a pair of actions. Unlike conventional online learning settings that require absolute feedback for each action, the dueling bandits framework assumes only the presence of (noisy) binary feedback about the relative quality of each pair of actions. The dueling bandits problem is well-suited for modeling settings that elicit subjective or implicit human feedback, which is typically more reliable in preference form. In this survey, we review recent results in the theories, algorithms, and applications of the dueling bandits problem. As an emerging domain, the theories and algorithms of dueling bandits have been intensively studied during the past few years. We provide an overview of recent advancements, including algorithmic advances and applications. We discuss extensions to standard problem formulation and novel application areas, highlighting key open research questions in our discussion.

Unlike traditional learning to rank models that depend on hand-crafted features, neural representation learning models learn higher level features for the ranking task by training on large datasets. Their ability to learn new features directly from the data, however, may come at a price. Without any special supervision, these models learn relationships that may hold only in the domain from which the training data is sampled, and generalize poorly to domains not observed during training. We study the effectiveness of adversarial learning as a cross domain regularizer in the context of the ranking task. We use an adversarial discriminator and train our neural ranking model on a small set of domains. The discriminator provides a negative feedback signal to discourage the model from learning domain specific representations. Our experiments show consistently better performance on held out domains in the presence of the adversarial discriminator---sometimes up to 30% on [email protected]$.

We address the problem of learning behaviour policies to optimise online metrics from heterogeneous usage data. While online metrics, e.g., click-through rate, can be optimised effectively using exploration data, such data is costly to collect in practice, as it temporarily degrades the user experience. Leveraging related data sources to improve online performance would be extremely valuable, but is not possible using current approaches. We formulate this task as a policy transfer learning problem, and propose a first solution, called collective noise contrastive estimation (collective NCE). NCE is an efficient solution to approximating the gradient of a log-softmax objective. Our approach jointly optimises embeddings of heterogeneous data to transfer knowledge from the source domain to the target domain. We demonstrate the effectiveness of our approach by learning an effective policy for an online radio station jointly from user-generated playlists, and usage data collected in an exploration bucket.

People often ask others for restaurant recommendations as a way to discover new dining experiences. This makes restaurant recommendation an exciting scenario for recommender systems and has led to substantial research in this area. However, most such systems behave very differently from a human when asked for a recommendation. The goal of this paper is to begin to reduce this gap. In particular, humans can quickly establish preferences when asked to make a recommendation for someone they do not know. We address this cold-start recommendation problem in an online learning setting. We develop a preference elicitation framework to identify which questions to ask a new user to quickly learn their preferences. Taking advantage of latent structure in the recommendation space using a probabilistic latent factor model, our experiments with both synthetic and real world data compare different types of feedback and question selection strategies. We find that our framework can make very effective use of online user feedback, improving personalized recommendations over a static model by 25% after asking only 2 questions. Our results demonstrate dramatic benefits of starting from offline embeddings, and highlight the benefit of bandit-based explore-exploit strategies in this setting.

We consider the problem of learning to choose actions using contextual information when provided with limited feedback in the form of relative pairwise comparisons. We study this problem in the dueling-bandits framework of Yue et al. (COLT’09), which we extend to incorporate context. Roughly, the learner’s goal is to find the best policy, or way of behaving, in some space of policies, although “best” is not always so clearly defined. Here, we propose a new and natural solution concept, rooted in game theory, called a \emphvon Neumann winner, a randomized policy that beats or ties every other policy. We show that this notion overcomes important limitations of existing solutions, particularly the Condorcet winner which has typically been used in the past, but which requires strong and often unrealistic assumptions. We then present three \emphefficient algorithms for online learning in our setting, and for approximating a von Neumann winner from batch-like data. The first of these algorithms achieves particularly low regret, even when data is adversarial, although its time and space requirements are linear in the size of the policy space. The other two algorithms require time and space only logarithmic in the size of the policy space when provided access to an oracle for solving classification problems on the space.

The gold standard for online retrieval evaluation is AB testing. Rooted in the idea of a controlled experiment, AB tests compare the performance of an experimental system (treatment) on one sample of the user population, to that of a baseline system (control) on another sample. Given an online evaluation metric that accurately reflects user satisfaction, these tests enjoy high validity. However, due to the high variance across users, these comparisons often have low sensitivity, requiring millions of queries to detect statistically significant differences between systems. Interleaving is an alternative online evaluation approach, where each user is presented with a combination of results from both the control and treatment systems. Compared to AB tests, interleaving has been shown to be substantially more sensitive. However, interleaving methods have so far focused on user clicks only, and lack support for more sophisticated user satisfaction metrics as used in AB testing. In this paper we present the first method for integrating user satisfaction metrics with interleaving. We show how interleaving can be extended to (1) directly match user signals and parameters of AB metrics, and (2) how parameterized interleaving credit functions can be automatically calibrated to predict AB outcomes. We also develop a new method for estimating the relative sensitivity of interleaving and AB metrics, and show that our interleaving credit functions improve agreement with AB metrics without sacrificing sensitivity. Our results, using 38 large-scale online experiments en- compassing over 3 billion clicks in a web search setting, demonstrate up to a 22% improvement in agreement with AB metrics (constituting over a 50% error reduction), while maintaining sensitivity of one to two orders of magnitude above the AB tests. This paves the way towards more sensitive and accurate online evaluation.

Query Auto Completion (QAC) suggests possible queries to web search users from the moment they start entering a query. This popular feature of web search engines is thought to reduce physical and cognitive effort when formulating a query.

Query Auto-Completion (QAC) is a popular feature of web search engines that aims to assist users to formulate queries faster and avoid spelling mistakes by presenting them with possible completions as soon as they start typing. However, despite the wide adoption of auto-completion in search systems, there is little published on how users interact with such services.

A result page of a modern web search engine is often much more complicated than a simple list of "ten blue links." In particular, a search engine may combine results from different sources (e.g., Web, News, and Images), and display these as grouped results to provide a better user experience. Such a system is called an aggregated or federated search system.

Online learning to rank for information retrieval (IR) holds promise for allowing the development of "self-learning" search engines that can automatically adjust to their users. With the large amount of e.g., click data that can be collected in web search settings, such techniques could enable highly scalable ranking optimization. However, feedback obtained from user interactions is noisy, and developing approaches that can learn from this feedback quickly and reliably is a major challenge.

As retrieval systems become more complex, learning to rank approaches are being developed to automatically tune their parameters. Using online learning to rank, retrieval systems can learn directly from implicit feedback inferred from user interactions. In such an online setting, algorithms must obtain feedback for effective learning while simultaneously utilizing what has already been learned to produce high quality results. We formulate this challenge as an exploration–exploitation dilemma and propose two methods for addressing it. By adding mechanisms for balancing exploration and exploitation during learning, each method extends a state-of-the-art learning to rank method, one based on listwise learning and the other on pairwise learning. Using a recently developed simulation framework that allows assessment of online performance, we empirically evaluate both methods. Our results show that balancing exploration and exploitation can substantially and significantly improve the online retrieval performance of both listwise and pairwise approaches. In addition, the results demonstrate that such a balance affects the two approaches in different ways, especially when user feedback is noisy, yielding new insights relevant to making online learning to rank effective in practice.

Ranker evaluation is central to the research into search engines, be it to compare rankers or to provide feedback for learning to rank. Traditional evaluation approaches do not scale well because they require explicit relevance judgments of document-query pairs, which are expensive to obtain. A promising alternative is the use of interleaved comparison methods, which compare rankers using click data obtained when interleaving their rankings. In this article, we propose a framework for analyzing interleaved comparison methods. An interleaved comparison method has fidelity if the expected outcome of ranker comparisons properly corresponds to the true relevance of the ranked documents. It is sound if its estimates of that expected outcome are unbiased and consistent. It is efficient if those estimates are accurate with only little data. We analyze existing interleaved comparison methods and find that, while sound, none meet our criteria for fidelity. We propose a probabilistic interleave method, which is sound and has fidelity. We show empirically that, by marginalizing out variables that are known, it is more efficient than existing interleaved comparison methods. Using importance sampling we derive a sound extension that is able to reuse historical data collected in previous comparisons of other ranker pairs.

Information retrieval evaluation most often involves manually assessing the relevance of particular query-document pairs. In cases where this is difficult (such as personalized search), interleaved comparison methods are becoming increasingly common. These methods compare pairs of ranking functions based on user clicks on search results, thus better reflecting true user preferences. However, by depending on clicks, there is a potential for bias. For example, users have been previously shown to be more likely to click on results with attractive titles and snippets. An interleaving evaluation where one ranker tends to generate results that attract more clicks (without being more relevant) may thus be biased.

Interleaved comparison methods, which compare rankers using click data, are a promising alternative to traditional information retrieval evaluation methods that require expensive explicit judgments. A major limitation of these methods is that they assume access to live data, meaning that new data must be collected for every pair of rankers compared. We investigate the use of previously collected click data (i.e., historical data) for interleaved comparisons. We start by analyzing to what degree existing interleaved comparison methods can be applied and find that a recent probabilistic method allows such data reuse, even though it is biased when applied to historical data. We then propose an interleaved comparison method that is based on the probabilistic approach but uses importance sampling to compensate for bias. We experimentally confirm that probabilistic methods make the use of historical data for interleaved comparisons possible and effective.

Evaluating rankers using implicit feedback, such as clicks on documents in a result list, is an increasingly popular alternative to traditional evaluation methods based on explicit relevance judgments. Previous work has shown that so-called interleaved comparison methods can utilize click data to detect small differences between rankers and can be applied to learn ranking functions online. In this paper, we analyze three existing interleaved comparison methods and find that they are all either biased or insensitive to some differences between rankers. To address these problems, we present a new method based on a probabilistic interleaving process. We derive an unbiased estimator of comparison outcomes and show how marginalizing over possible comparison outcomes given the observed click data can make this estimator even more effective.

The goal of my research is to develop self-learning search engines, that can learn online, i.e., directly from interactions with actual users. Such systems can continuously adapt to user preferences throughout their lifetime, leading to better search performance in settings where expensive manual tuning is infeasible. Challenges that are addressed in my work include the development of effective online learning to rank algorithms for IR, user aspects, and evaluation.

Traditional retrieval evaluation uses explicit relevance judgments which are expensive to collect. Relevance assessments inferred from implicit feedback such as click-through data can be collected inexpensively, but may be less reliable. We compare assessments derived from click-through data to another source of implicit feedback that we assume to be highly indicative of relevance: purchase decisions. Evaluating retrieval runs based on a log of an audio-visual archive, we find agreement between system rankings and purchase decisions to be surprisingly high.

Abstract Expertise‐seeking research studies how people search for expertise and choose whom to contact in the context of a specific task. An important outcome are models that identify factors that influence expert finding. Expertise retrieval addresses the same problem, expert finding, but from a system‐centered perspective. The main focus has been on developing content‐based algorithms similar to document search. These algorithms identify matching experts primarily on the basis of the textual content of documents with which experts are associated. Other factors, such as the ones identified by expertise‐seeking models, are rarely taken into account. In this article, we extend content‐based expert‐finding approaches with contextual factors that have been found to influence human expert finding. We focus on a task of science communicators in a knowledge‐intensive environment, the task offinding similar experts, given an example expert. Our approach combines expertise‐seeking and retrieval research. First, we conduct a user study to identify contextual factors that may play a role in the studied task and environment. Then, we design expert retrieval models to capture these factors. We combine these with content‐based retrieval models and evaluate them in a retrieval experiment. Our main finding is that while content‐based features are the most important, human participants also take contextual factors into account, such as media experience and organizational structure. We develop two principled ways of modeling the identified factors and integrate them with content‐based retrieval models. Our experiments show that models combining content‐based and contextual factors can significantly outperform existing content‐based models.

Keyphrases are short phrases that reflect the main topic of a document. Because manually annotating documents with keyphrases is a time-consuming process, several automatic approaches have been developed. Typically, candidate phrases are extracted using features such as position or frequency in the document text. Document structure may contain useful information about which parts or phrases of a document are important, but has rarely been considered as a source of information for keyphrase extraction.

Lexical taxonomies, such as WordNet, are important resources underlying natural language processing techniques such as machine translation and word-sense disambiguation. However, creating and maintaining such taxonomies manually is a tedious and time-consuming task. This has led to a great deal of interest in automatic methods for retrieving taxonomic relations. Efforts for both manual development of taxonomies and automatic acquisition methods have largely focused on English-language resources. Although WordNets exist for many languages, these are usually much smaller than Princeton WordNet (PWN) [1], the major semantic resource for English. For English, an excellent method has been developed for automatically extending lexical taxonomies. Snow et al. [4] show that it is possible to predict new and precise hypernymhyponym relations from a parsed corpus. Unfortunately, the method cannot be easily applied to other languages, as it relies on existing lexical resources. First, the authors use the large amount of data already contained in the PWN to extract a pattern lexicon and train their hypernym classifier. Second, the authors apply word-sense disambiguation based on an existing sense-tagged corpus. Problems in transferring the method to another language include the small size of non-English WordNets and the lack of sense-tagged corpora in other languages. In this paper we apply the method of [4] to a non-English language (Dutch) for which only a basic WordNet and a parser are available. We find that, without an additional sense-tagged corpus, this approach is highly susceptible to noise due to word sense ambiguity. We propose and evaluate two methods to address this problem.

We propose a technique for user modeling in Adaptive Hypermedia (AH) that is unobtrusive at both the level of observable behavior and that of cognition. Unobtrusive user modeling is complementary to transparent user modeling. Unobtrusive user modeling induces user models appropriate for Educational AH (EAH) based on metaphors characterized by a shared locus of control over learning, such as small group learning. Transparent user modeling, on the other hand, induces user models appropriate for EAH based on metaphors characterized by an independent locus of control, such as social navigation. We extend an existing decomposition model of adaptation in AH to incorporate summative evaluation of unobtrusive user modeling. Summative evaluation examines the effects or outcomes of a system, while formative evaluation is used to improve a system under development. We also introduce a separate, two-stage model for formative evaluation of unobtrusive user modeling. We then present results from a field study for the first of the two developmental stages and describe the current field study of the second developmental stage. Finally, we compare unobtrusive and transparent user modeling, and explain the role of each.