Message passing neural networks have shown a lot of success on graph-structured data. However, there are many instances where message passing can lead to over-smoothing or fail when neighboring nodes belong to different classes. In this work, we introduce a simple yet general framework for improving learning in message passing neural networks. Our approach essentially upsamples edges in the original graph by adding "slow nodes" at each edge that can mediate communication between a source and a target node. Our method only modifies the input graph, making it plug-and-play and easy to use with existing models. To understand the benefits of slowing down message passing, we provide theoretical and empirical analyses. We report results on several supervised and self-supervised benchmarks, and show improvements across the board, notably in heterophilic conditions where adjacent nodes are more likely to have different labels. Finally, we show how our approach can be used to generate augmentations for self-supervised learning, where slow nodes are randomly introduced into different edges in the graph to generate multi-scale views with variable path lengths.

Representation learning and exploration are among the key challenges for any deep reinforcement learning agent. In this work, we provide a singular value decomposition based method that can be used to obtain representations that preserve the underlying transition structure in the domain. Perhaps interestingly, we show that these representations also capture the relative frequency of state visitations, thereby providing an estimate for pseudo-counts for free. To scale this decomposition method to large-scale domains, we provide an algorithm that never requires building the transition matrix, can make use of deep networks, and also permits mini-batch training. Further, we draw inspiration from predictive state representations and extend our decomposition method to partially observable environments. With experiments on multi-task settings with partially observable domains, we show that the proposed method can not only learn useful representation on DM-Lab-30 environments (that have inputs involving language instructions, pixel images, rewards, among others) but it can also be effective at hard exploration tasks in DM-Hard-8 environments.

We study the learning dynamics of self-predictive learning for reinforcement learning, a family of algorithms that learn representations by minimizing the prediction error of their own future latent representations. Despite its recent empirical success, such algorithms have an apparent defect: trivial representations (such as constants) minimize the prediction error, yet it is obviously undesirable to converge to such solutions. Our central insight is that careful designs of the optimization dynamics are critical to learning meaningful representations. We identify that a faster paced optimization of the predictor and semi-gradient updates on the representation, are crucial to preventing the representation collapse. Then in an idealized setup, we show self-predictive learning dynamics carries out spectral decomposition on the state transition matrix, effectively capturing information of the transition dynamics. Building on the theoretical insights, we propose bidirectional self-predictive learning, a novel self-predictive algorithm that learns two representations simultaneously. We examine the robustness of our theoretical insights with a number of small-scale experiments and showcase the promise of the novel representation learning algorithm with large-scale experiments.

Unconstrained and natural behavior consists of dynamics that are complex and unpredictable, especially when trying to predict what will happen multiple steps into the future. While some success has been found in building representations of animal behavior under constrained or simplified task-based conditions, many of these models cannot be applied to free and naturalistic settings where behavior becomes increasingly hard to model. In this work, we develop a multi-task representation learning model for animal behavior that combines two novel components: (i) an action-prediction objective that aims to predict the distribution of actions over future timesteps, and (ii) a multi-scale architecture that builds separate latent spaces to accommodate short- and long-term dynamics. After demonstrating the ability of the method to build representations of both local and global dynamics in robots in varying environments and terrains, we apply our method to the MABe 2022 Multi-Agent Behavior challenge, where our model ranks first overall on both mice and fly benchmarks. In all of these cases, we show that our model can build representations that capture the many different factors that drive behavior and solve a wide range of downstream tasks.

We introduce a method for policy improvement that interpolates between the greedy approach of value-based reinforcement learning (RL) and the full planning approach typical of model-based RL. The new method builds on the concept of a geometric horizon model (GHM, also known as a \gamma-model), which models the discounted state-visitation distribution of a given policy. We show that we can evaluate any non-Markov policy that switches between a set of base Markov policies with fixed probability by a careful composition of the base policy GHMs, without any additional learning. We can then apply generalised policy improvement (GPI) to collections of such non-Markov policies to obtain a new Markov policy that will in general outperform its precursors. We provide a thorough theoretical analysis of this approach, develop applications to transfer and standard RL, and empirically demonstrate its effectiveness over standard GPI on a challenging deep RL continuous control task. We also provide an analysis of GHM training methods, proving a novel convergence result regarding previously proposed methods and showing how to train these models stably in deep RL settings.

We present BYOL-Explore, a conceptually simple yet general approach for curiosity-driven exploration in visually complex environments. BYOL-Explore learns the world representation, the world dynamics and the exploration policy all-together by optimizing a single prediction loss in the latent space with no additional auxiliary objective. We show that BYOL-Explore is effective in DM-HARD-8, a challenging partially-observable continuous-action hard-exploration benchmark with visually rich 3-D environment. On this benchmark, we solve the majority of the tasks purely through augmenting the extrinsic reward with BYOL-Explore intrinsic reward, whereas prior work could only get off the ground with human demonstrations. As further evidence of the generality of BYOL-Explore, we show that it achieves superhuman performance on the ten hardest exploration games in Atari while having a much simpler design than other competitive agents.

Credit assignment in reinforcement learning is the problem of measuring an action’s influence on future rewards. In particular, this requires separating skill from luck, i.e. disentangling the effect of an action on rewards from that of external factors and subsequent actions. To achieve this, we adapt the notion of counterfactuals from causality theory to a model-free RL setup. The key idea is to condition value functions on future events, by learning to extract relevant information from a trajectory. We formulate a family of policy gradient algorithms that use these future-conditional value functions as baselines or critics, and show that they are provably low variance. To avoid the potential bias from conditioning on future information, we constrain the hindsight information to not contain information about the agent’s actions. We demonstrate the efficacy and validity of our algorithm on a number of illustrative and challenging problems.

Deep neural networks are revolutionizing the way complex systems are designed. Consequently, there is a pressing need for tools and techniques for network analysis and certification. To help in addressing that need, we present Marabou, a framework for verifying deep neural networks. Marabou is an SMT-based tool that can answer queries about a network’s properties by transforming these queries into constraint satisfaction problems. It can accommodate networks with different activation functions and topologies, and it performs high-level reasoning on the network that can curtail the search space and improve performance. It also supports parallel execution to further enhance scalability. Marabou accepts multiple input formats, including protocol buffer files generated by the popular TensorFlow framework for neural networks. We describe the system architecture and main components, evaluate the technique and discuss ongoing work.

We describe MultiSynth, a framework for synthesizing domain-specific programs from a multimodal dataset of examples. Given a domain-specific language (DSL), a dataset is multimodal if there is no single program in the DSL that generalizes over all the examples. Further, even if the examples in the dataset were generalized in terms of a set of programs, the domains of these programs may not be disjoint, thereby leading to ambiguity in synthesis. MultiSynth is a framework that incorporates concepts of synthesizing programs with minimum generality, while addressing the need of accurate prediction. We show how these can be achieved through (i) transformation driven partitioning of the dataset, (ii) least general generalization, for a generalized specification of the input and the output, and (iii) learning to rank, for estimating feature weights in order to map an input to the most appropriate mode in case of ambiguity. We show the effectiveness of our framework in two domains: in the first case, we extend an existing approach for synthesizing programs for XML tree transformations to ambiguous multimodal datasets. In the second case, MultiSynth is used to preorder words for machine translation, by learning permutations of productions in the parse trees of the source side sentences. Our evaluations reflect the effectiveness of our approach.