Showing papers by "Matthew Botvinick published in 2021"

Flexible modulation of sequence generation in the entorhinal-hippocampal system

Abstract: Exploration, consolidation and planning depend on the generation of sequential state representations. However, these algorithms require disparate forms of sampling dynamics for optimal performance. We theorize how the brain should adapt internally generated sequences for particular cognitive functions and propose a neural mechanism by which this may be accomplished within the entorhinal-hippocampal circuit. Specifically, we demonstrate that the systematic modulation along the medial entorhinal cortex dorsoventral axis of grid population input into the hippocampus facilitates a flexible generative process that can interpolate between qualitatively distinct regimes of sequential hippocampal reactivations. By relating the emergent hippocampal activity patterns drawn from our model to empirical data, we explain and reconcile a diversity of recently observed, but apparently unrelated, phenomena such as generative cycling, diffusive hippocampal reactivations and jumping trajectory events.

Unsupervised deep learning identifies semantic disentanglement in single inferotemporal face patch neurons.

Abstract: In order to better understand how the brain perceives faces, it is important to know what objective drives learning in the ventral visual stream. To answer this question, we model neural responses to faces in the macaque inferotemporal (IT) cortex with a deep self-supervised generative model, β-VAE, which disentangles sensory data into interpretable latent factors, such as gender or age. Our results demonstrate a strong correspondence between the generative factors discovered by β-VAE and those coded by single IT neurons, beyond that found for the baselines, including the handcrafted state-of-the-art model of face perception, the Active Appearance Model, and deep classifiers. Moreover, β-VAE is able to reconstruct novel face images using signals from just a handful of cells. Together our results imply that optimising the disentangling objective leads to representations that closely resemble those in the IT at the single unit level. This points at disentangling as a plausible learning objective for the visual brain. Little is known about the brain’s computations that enable the recognition of faces. Here, the authors use unsupervised deep learning to show that the brain disentangles faces into semantically meaningful factors, like age or the presence of a smile, at the single neuron level.

A rapid and efficient learning rule for biological neural circuits

Abstract: The dominant view in neuroscience is that changes in synaptic weights underlie learning. It is unclear, however, how the brain is able to determine which synapses should change, and by how much. This uncertainty stands in sharp contrast to deep learning, where changes in weights are explicitly engineered to optimize performance. However, the main tool for doing that, backpropagation, is not biologically plausible, and networks trained with this rule tend to forget old tasks when learning new ones. Here we introduce the Dendritic Gated Network (DGN), a variant of the Gated Linear Network, which offers a biologically plausible alternative to backpropagation. DGNs combine dendritic "gating" (whereby interneurons target dendrites to shape neuronal response) with local learning rules to yield provably efficient performance. They are significantly more data efficient than conventional artificial networks and are highly resistant to forgetting, and we show that they perform well on a variety of tasks, in some cases better than backpropagation. The DGN bears similarities to the cerebellum, where there is evidence for shaping of Purkinje cell responses by interneurons. It also makes several experimental predictions, one of which we validate with in vivo cerebellar imaging of mice performing a motor task.

Neural evidence for the successor representation in choice evaluation

Abstract: Evaluating choices in multi-step tasks is thought to involve mentally simulating trajectories. Recent theories propose that the brain simplifies these laborious computations using temporal abstraction: storing actions' consequences, collapsed over multiple timesteps (the Successor Representation; SR). Although predictive neural representations and, separately, behavioral errors ("slips of action") consistent with this mechanism have been reported, it is unknown whether these neural representations support choices in a manner consistent with the SR. We addressed this question by using fMRI to measure predictive representations in a setting where the SR implies specific errors in multi-step expectancies and corresponding behavioral errors. By decoding measures of state predictions from sensory cortex during choice evaluation, we identified evidence that behavioral errors predicted by the SR are accompanied by predictive representations of upcoming task states reflecting SR predicted erroneous multi-step expectancies. These results provide neural evidence for the SR in choice evaluation and contribute toward a mechanistic understanding of flexible and inflexible decision making.

Generative replay for compositional visual understanding in the prefrontal-hippocampal circuit

Abstract: Understanding the visual world is a constructive process. Whilst a frontal-hippocampal circuit is known to be essential for this task, little is known about the associated neuronal computations. Visual understanding appears superficially distinct from other known functions of this circuit, such as spatial reasoning and model-based planning, but recent models suggest deeper computational similarities. Here, using fMRI, we show that representations of a simple visual scene in these brain regions are relational and compositional – key computational properties theorised to support rapid construction of hippocampal maps. Using MEG, we show that rapid sequences of representations, akin to replay in spatial navigation and planning problems, are also engaged in visual construction. Whilst these sequences have previously been proposed as mechanisms to plan possible futures or learn from the past, here they are used to understand the present. Replay sequences form constructive hypotheses about possible scene configurations. These hypotheses play out in an optimal order for relational inference, progressing from predictable to uncertain scene elements, gradually constraining possible configurations, and converging on the correct scene configuration. Together, these results suggest a computational bridge between apparently distinct functions of hippocampal-prefrontal circuitry, and a role for generative replay in constructive inference and hypothesis testing.

Perceiver IO: A General Architecture for Structured Inputs & Outputs

Abstract: The recently-proposed Perceiver model obtains good results on several domains (images, audio, multimodal, point clouds) while scaling linearly in compute and memory with the input size. While the Perceiver supports many kinds of inputs, it can only produce very simple outputs such as class scores. Perceiver IO overcomes this limitation without sacrificing the original's appealing properties by learning to flexibly query the model's latent space to produce outputs of arbitrary size and semantics. Perceiver IO still decouples model depth from data size and still scales linearly with data size, but now with respect to both input and output sizes. The full Perceiver IO model achieves strong results on tasks with highly structured output spaces, such as natural language and visual understanding, StarCraft II, and multi-task and multi-modal domains. As highlights, Perceiver IO matches a Transformer-based BERT baseline on the GLUE language benchmark without the need for input tokenization and achieves state-of-the-art performance on Sintel optical flow estimation.

Alchemy: A structured task distribution for meta-reinforcement learning.

Abstract: There has been rapidly growing interest in meta-learning as a method for increasing the flexibility and sample efficiency of reinforcement learning One problem in this area of research, however, has been a scarcity of adequate benchmark tasks In general, the structure underlying past benchmarks has either been too simple to be inherently interesting, or too ill-defined to support principled analysis In the present work, we introduce a new benchmark for meta-RL research, which combines structural richness with structural transparency Alchemy is a 3D video game, implemented in Unity, which involves a latent causal structure that is resampled procedurally from episode to episode, affording structure learning, online inference, hypothesis testing and action sequencing based on abstract domain knowledge We evaluate a pair of powerful RL agents on Alchemy and present an in-depth analysis of one of these agents Results clearly indicate a frank and specific failure of meta-learning, providing validation for Alchemy as a challenging benchmark for meta-RL Concurrent with this report, we are releasing Alchemy as public resource, together with a suite of analysis tools and sample agent trajectories

Alchemy: A benchmark and analysis toolkit for meta-reinforcement learning agents

Abstract: There has been rapidly growing interest in meta-learning as a method for increasing the flexibility and sample efficiency of reinforcement learning. One problem in this area of research, however, has been a scarcity of adequate benchmark tasks. In general, the structure underlying past benchmarks has either been too simple to be inherently interesting, or too ill-defined to support principled analysis. In the present work, we introduce a new benchmark for meta-RL research, emphasizing transparency and potential for in-depth analysis as well as structural richness. Alchemy is a 3D video game, implemented in Unity, which involves a latent causal structure that is resampled procedurally from episode to episode, affording structure learning, online inference, hypothesis testing and action sequencing based on abstract domain knowledge. We evaluate a pair of powerful RL agents on Alchemy and present an in-depth analysis of one of these agents. Results clearly indicate a frank and specific failure of meta-learning, providing validation for Alchemy as a challenging benchmark for meta-RL. Concurrent with this report, we are releasing Alchemy as public resource, together with a suite of analysis tools and sample agent trajectories.

Synthetic Returns for Long-Term Credit Assignment.

Abstract: Since the earliest days of reinforcement learning, the workhorse method for assigning credit to actions over time has been temporal-difference (TD) learning, which propagates credit backward timestep-by-timestep. This approach suffers when delays between actions and rewards are long and when intervening unrelated events contribute variance to long-term returns. We propose state-associative (SA) learning, where the agent learns associations between states and arbitrarily distant future rewards, then propagates credit directly between the two. In this work, we use SA-learning to model the contribution of past states to the current reward. With this model we can predict each state's contribution to the far future, a quantity we call "synthetic returns". TD-learning can then be applied to select actions that maximize these synthetic returns (SRs). We demonstrate the effectiveness of augmenting agents with SRs across a range of tasks on which TD-learning alone fails. We show that the learned SRs are interpretable: they spike for states that occur after critical actions are taken. Finally, we show that our IMPALA-based SR agent solves Atari Skiing -- a game with a lengthy reward delay that posed a major hurdle to deep-RL agents -- 25 times faster than the published state-of-the-art.

SIMONe: View-Invariant, Temporally-Abstracted Object Representations via Unsupervised Video Decomposition

Abstract: To help agents reason about scenes in terms of their building blocks, we wish to extract the compositional structure of any given scene (in particular, the configuration and characteristics of objects comprising the scene). This problem is especially difficult when scene structure needs to be inferred while also estimating the agent's location/viewpoint, as the two variables jointly give rise to the agent's observations. We present an unsupervised variational approach to this problem. Leveraging the shared structure that exists across different scenes, our model learns to infer two sets of latent representations from RGB video input alone: a set of "object" latents, corresponding to the time-invariant, object-level contents of the scene, as well as a set of "frame" latents, corresponding to global time-varying elements such as viewpoint. This factorization of latents allows our model, SIMONe, to represent object attributes in an allocentric manner which does not depend on viewpoint. Moreover, it allows us to disentangle object dynamics and summarize their trajectories as time-abstracted, view-invariant, per-object properties. We demonstrate these capabilities, as well as the model's performance in terms of view synthesis and instance segmentation, across three procedurally generated video datasets.

Rapid Task-Solving in Novel Environments

Abstract: We propose the challenge of rapid task-solving in novel environments (RTS), wherein an agent must solve a series of tasks as rapidly as possible in an unfamiliar environment. An effective RTS agent must balance between exploring the unfamiliar environment and solving its current task, all while building a model of the new environment over which it can plan when faced with later tasks. While modern deep RL agents exhibit some of these abilities in isolation, none are suitable for the full RTS challenge. To enable progress toward RTS, we introduce two challenge domains: (1) a minimal RTS challenge called the Memory\&Planning Game and (2) One-Shot StreetLearn Navigation, which introduces scale and complexity from real-world data. We demonstrate that state-of-the-art deep RL agents fail at RTS in both domains, and that this failure is due to an inability to plan over gathered knowledge. We develop Episodic Planning Networks (EPNs) and show that deep-RL agents with EPNs excel at RTS, outperforming the nearest baseline by factors of 2-3 and learning to navigate held-out StreetLearn maps within a single episode. We show that EPNs learn to execute a value iteration-like planning algorithm and that they generalize to situations beyond their training experience.

Attention over Learned Object Embeddings Enables Complex Visual Reasoning

Abstract: Neural networks have achieved success in a wide array of perceptual tasks but often fail at tasks involving both perception and higher-level reasoning. On these more challenging tasks, bespoke approaches (such as modular symbolic components, independent dynamics models or semantic parsers) targeted towards that specific type of task have typically performed better. The downside to these targeted approaches, however, is that they can be more brittle than general-purpose neural networks, requiring significant modification or even redesign according to the particular task at hand. Here, we propose a more general neural-network-based approach to dynamic visual reasoning problems that obtains state-of-the-art performance on three different domains, in each case outperforming bespoke modular approaches tailored specifically to the task. Our method relies on learned object-centric representations, self-attention and self-supervised dynamics learning, and all three elements together are required for strong performance to emerge. The success of this combination suggests that there may be no need to trade off flexibility for performance on problems involving spatio-temporal or causal-style reasoning. With the right soft biases and learning objectives in a neural network we may be able to attain the best of both worlds.

Deep reinforcement learning models the emergent dynamics of human cooperation

Abstract: Collective action demands that individuals efficiently coordinate how much, where, and when to cooperate. Laboratory experiments have extensively explored the first part of this process, demonstrating that a variety of social-cognitive mechanisms influence how much individuals choose to invest in group efforts. However, experimental research has been unable to shed light on how social cognitive mechanisms contribute to the where and when of collective action. We leverage multi-agent deep reinforcement learning to model how a social-cognitive mechanism--specifically, the intrinsic motivation to achieve a good reputation--steers group behavior toward specific spatial and temporal strategies for collective action in a social dilemma. We also collect behavioral data from groups of human participants challenged with the same dilemma. The model accurately predicts spatial and temporal patterns of group behavior: in this public goods dilemma, the intrinsic motivation for reputation catalyzes the development of a non-territorial, turn-taking strategy to coordinate collective action.

Collaborating with Humans without Human Data

Abstract: Collaborating with humans requires rapidly adapting to their individual strengths, weaknesses, and preferences. Unfortunately, most standard multi-agent reinforcement learning techniques, such as self-play (SP) or population play (PP), produce agents that overfit to their training partners and do not generalize well to humans. Alternatively, researchers can collect human data, train a human model using behavioral cloning, and then use that model to train "human-aware" agents ("behavioral cloning play", or BCP). While such an approach can improve the generalization of agents to new human co-players, it involves the onerous and expensive step of collecting large amounts of human data first. Here, we study the problem of how to train agents that collaborate well with human partners without using human data. We argue that the crux of the problem is to produce a diverse set of training partners. Drawing inspiration from successful multi-agent approaches in competitive domains, we find that a surprisingly simple approach is highly effective. We train our agent partner as the best response to a population of self-play agents and their past checkpoints taken throughout training, a method we call Fictitious Co-Play (FCP). Our experiments focus on a two-player collaborative cooking simulator that has recently been proposed as a challenge problem for coordination with humans. We find that FCP agents score significantly higher than SP, PP, and BCP when paired with novel agent and human partners. Furthermore, humans also report a strong subjective preference to partnering with FCP agents over all baselines.

Pre-trial predictors of conflict response efficacy in human dorsolateral prefrontal cortex

Abstract: The ability to perform motor actions depends, in part, on the brain9s initial state, that is the ensemble firing rate pattern prior to the initiation of action. We hypothesized that the same principle would apply to cognitive functions as well. To test this idea, we examined a unique set of single unit data collected in human dorsolateral prefrontal cortex (dlPFC). Data were collected in a conflict task that interleaves Simon (motor-type) and Eriksen (flanker-type) conflict trials. Variability in pre-trial firing rate predicted the ability to resolve conflict, as inferred from reaction times. Ensemble patterns that predicted faster Simon reaction times overlapped slightly with those predicting Eriksen performance, indicating that the two conflict types are associated with near-orthogonal initial states, and suggesting that there is a weak abstract or amodal conflict preparatory state in this region. These codes became fully orthogonalized in the response state. We interpret these results in light of the initial state and dual-mechanisms of control hypotheses, arguing that the firing patterns in dlPFC immediately preceding the start of the task predispose it for the efficient implementation of cognitive action.