CURL uses contrastive learning to improve reinforcement learning performance.
problem Improving reinforcement learning performance on complex tasks.
method Contrastive learning to extract high-level features from raw pixels, followed by off-policy control.
result CURL outperforms prior methods on DeepMind Control Suite and Atari Games.
Deep RL algorithms can overfit to early experiences, leading to poor performance.
problem Overfitting to early interactions in deep reinforcement learning.
method Proposed a mechanism to periodically reset part of the agent to mitigate overfitting.
result Periodic resetting improves performance in both discrete and continuous action domains.
Learning to control an environment without hand-crafted rewards or expert data remains challenging and is at the frontier of reinforcement learning research. We present an unsupervised learning algorithm to train agents to achieve perceptually-specified goals using only a stream of observations and actions. Our agent s…
New method QMLE performs well in complex action spaces without policy gradients.
problem Why policy gradients outperform action-value methods in complex action spaces.
method QMLE framework for action-value methods based on three principles.
result QMLE performs comparably to policy gradient methods in complex action spaces.
RAD enhances RL algorithms with data augmentations.
problem Challenges in RL learning from visual observations.
method RAD is a simple plug-and-play module for RL algorithms.
result RAD improves data-efficiency and final performance.
We revisit residual algorithms in both model-free and model-based reinforcement learning settings. We propose the bidirectional target network technique to stabilize residual algorithms, yielding a residual version of DDPG that significantly outperforms vanilla DDPG in the DeepMind Control Suite benchmark. Moreover, we…
Exploration in sparse reward reinforcement learning remains an open challenge. Many state-of-the-art methods use intrinsic motivation to complement the sparse extrinsic reward signal, giving the agent more opportunities to receive feedback during exploration. Commonly these signals are added as bonus rewards, which res…
This paper studies accelerations in Q-learning algorithms. We propose an accelerated target update scheme by incorporating the historical iterates of Q functions. The idea is conceptually inspired by the momentum-based accelerated methods in the optimization theory. Conditions under which the proposed accelerated algor…
We present an off-policy actor-critic algorithm for Reinforcement Learning (RL) that combines ideas from gradient-free optimization via stochastic search with learned action-value function. The result is a simple procedure consisting of three steps: i) policy evaluation by estimating a parametric action-value function;…
Invariances to translation, rotation and other spatial transformations are a hallmark of the laws of motion, and have widespread use in the natural sciences to reduce the dimensionality of systems of equations. In supervised learning, such as in image classification tasks, rotation, translation and scale invariances ar…
Curious Replay improves model-based reinforcement learning agents' adaptability.
problem Existing model-based reinforcement learning agents struggle to adapt quickly to changing environments.
method Curious Replay uses a curiosity-based priority signal for prioritized experience replay tailored to model-based agents.
result Agents using Curious Replay achieve improved performance in exploration and on benchmarks.
Simple image augmentation boosts deep RL from pixels.
problem Training deep RL models from raw pixel inputs is challenging.
method Applying standard computer vision data augmentation to reinforcement learning.
result Our method significantly improves SAC's performance on DeepMind control suite.
Experience replay is an important technique for addressing sample-inefficiency in deep reinforcement learning (RL), but faces difficulty in learning from binary and sparse rewards due to disproportionately few successful experiences in the replay buffer. Hindsight experience replay (HER) was recently proposed to tackle…
Self-supervised policy adapts after deployment without rewards.
problem Generalizing reinforcement learning policies across different environments.
method Uses self-supervision to train policies in new environments without reward signals.
result Significant improvements in generalization across diverse environments.
Proposes PIC and POIC for measuring task difficulty in RL.
problem Lack of metrics to measure task difficulty in RL.
method Introduces policy information capacity (PIC) and policy-optimal information capacity (POIC) as metrics based on mutual information.
result Empirically shows PIC and POIC correlate with task solvability better than alternatives.
This paper introduces the Behaviour Suite for Reinforcement Learning, or bsuite for short. bsuite is a collection of carefully-designed experiments that investigate core capabilities of reinforcement learning (RL) agents with two objectives. First, to collect clear, informative and scalable problems that capture key is…
The paper identifies overfitting as the main bottleneck in efficient deep reinforcement learning.
problem Improving sample efficiency in deep reinforcement learning.
method Empirical analysis on DMC tasks to identify overfitting as the main issue and developing a hill-climbing method targeting validation TD error.
result Overfitting is the primary bottleneck in sample-efficient deep RL, and regularization techniques can control this.
MAPS and MAPS-SE learn from multiple suboptimal experts to improve policies efficiently.
problem Learning from multiple suboptimal experts in reinforcement learning.
method Active policy improvement from multiple black-box oracles.
result MAPS and MAPS-SE achieve sample efficiency and state-wise imitation learning.
BCRL learns a Bellman complete representation for offline RL policy evaluation.
problem Learning a Q-function efficiently from offline data.
method BCRL learns a linear Bellman complete representation directly from data, enabling efficient OPE.
result BCRL achieves competitive OPE error and outperforms FQE in certain scenarios.
CUDC collects diverse data for offline RL by predicting future states.
problem Challenges in collecting task-agnostic data for offline RL.
method Adaptive temporal distances for curiosity-driven data collection.
result CUDC outperforms existing unsupervised methods in offline RL tasks.
Proposes method to discover diverse near-optimal policies in reinforcement learning.
problem Finding different solutions to the same problem in reinforcement learning.
method Formalizes problem as CMDP, uses Successor Features, proposes new diversity rewards.
result Proposed method discovers diverse near-optimal policies that are robust and distinct.
New method dynamically adjusts UTD ratio to balance under- and overfitting in RL.
problem Balancing under- and overfitting in world model learning for RL.
method Dynamic adjustment of UTD ratio based on validation performance on a small subset of experience data.
result Our method improves balance between under- and overfitting compared to default settings and competitive with extensive hyperparameter search.
CCLF improves sample efficiency in RL by prioritizing informative samples.
problem Learning from high-dimensional observations is challenging and unstable.
method Contrastive-Curiosity-Driven Learning Framework (CCLF) that prioritizes informative samples.
result CCLF improves sample efficiency and learning performance in RL.
MADE improves exploration in RL by maximizing deviation from explored regions.
problem Efficient exploration in high-dimensional RL tasks with sparse rewards.
method Proposes a new exploration approach via maximizing the deviation of the occupancy of the next policy from explored regions, adding it as an adaptive regularizer to the RL objective.
result Significantly improves sample efficiency in navigation and locomotion tasks.
Paper presents a world model that learns invariant causal features using contrastive unsupervised learning.
problem Learning invariant causal features in unsupervised settings.
method Contrastive unsupervised learning with intervention invariant auxiliary task.
result Significantly outperforms state-of-the-art methods on out-of-distribution point navigation tasks.
Deep RL evaluation underestimates uncertainty, leading to misleading conclusions.
problem Statistical uncertainty in deep RL performance evaluations is underestimated, leading to misleading conclusions.
method Advocates for reporting interval estimates of aggregate performance and proposes performance profiles to account for variability.
result Substantial discrepancies in prior performance comparisons are revealed, highlighting the need for more rigorous evaluation methods.
New unsupervised learning task improves RL performance.
problem Reward-driven feature learning limitations in RL from images.
method Introduce Augmented Temporal Contrast (ATC) for unsupervised learning of image representations.
result Training encoders using ATC matches or outperforms end-to-end RL in most environments.
Deep reinforcement learning (RL) agents often fail to generalize to unseen environments (yet semantically similar to trained agents), particularly when they are trained on high-dimensional state spaces, such as images. In this paper, we propose a simple technique to improve a generalization ability of deep RL agents by…
Deep Reinforcement Learning (DRL) has become increasingly powerful in recent years, with notable achievements such as Deepmind's AlphaGo. It has been successfully deployed in commercial vehicles like Mobileye's path planning system. However, a vast majority of work on DRL is focused on toy examples in controlled synthe…
We describe TF-Replicator, a framework for distributed machine learning designed for DeepMind researchers and implemented as an abstraction over TensorFlow. TF-Replicator simplifies writing data-parallel and model-parallel research code. The same models can be effortlessly deployed to different cluster architectures (i…
Deep GNNs and self-supervision boost graph learning at scale.
problem Efficiently deploying GNNs at large scale remains challenging.
method Two large-scale GNNs: a deep transductive node classifier and a very deep inductive graph regressor.
result Award-level performance on MAG240M and PCQM4M benchmarks.
In this paper a neural network heuristic dynamic programing (HDP) is used for optimal control of the virtual inertia based control of grid connected three phase inverters. It is shown that the conventional virtual inertia controllers are not suited for non inductive grids. A neural network based controller is proposed …
LLM Pro Finance Suite enhances financial NLP with instruction-tuned models.
problem Limited NLP capabilities for financial tasks in generalist models.
method Instruction-tuned large language models fine-tuned on financial data.
result Consistent improvement over state-of-the-art baselines in finance tasks.
Paper tackles overestimation bias in continuous control, improving performance by 25%.
problem Overestimation bias in off-policy learning.
method Truncated Quantile Critics (TQC) combines distributional representation, truncation, and ensembling of critics.
result TQC outperforms state-of-the-art methods by 25% on the Humanoid environment.
RL Unplugged benchmarks offline RL methods across diverse domains.
problem Evaluate offline reinforcement learning methods without online data collection.
method Proposes a benchmark suite with diverse datasets and detailed evaluation protocols.
result Demonstrates the effectiveness of offline RL methods across various domains.
Research explores how interconnected systems synchronize and how to control their behavior.
problem Understanding and controlling the behavior of interconnected dynamical systems.
method Mean field games approach applied to controlled coupled oscillators.
result Developed methods to predict and influence emergent phenomena in interconnected systems.
We approach the development of models and control strategies of susceptible-infected-susceptible (SIS) epidemic processes from the perspective of marked temporal point processes and stochastic optimal control of stochastic differential equations (SDEs) with jumps. In contrast to previous work, this novel perspective is…
Graph neural networks learn decentralized controllers from data.
problem Finding optimal decentralized controllers for autonomous agents is challenging.
method Adapting graph neural networks to handle delayed communications and ensure scalability and transferability.
result Graph neural networks can learn decentralized controllers from data, addressing the scalability and practical implementation issues of centralized controllers.
Paper establishes baselines for offline RL from visual observations.
problem Challenges in offline reinforcement learning from visual observations with continuous action spaces.
method Simple baselines and benchmarking tasks for offline RL from visual observations.
result Simple modifications to existing online RL algorithms outperform existing offline RL methods.
New method trains deep ResNets without normalization, achieving state-of-the-art performance.
problem Training deep ResNets without normalization layers leads to instability and lower accuracy.
method Adaptive gradient clipping and Normalizer-Free ResNets design.
result Normalizer-Free ResNets achieve 86.5% top-1 accuracy on ImageNet, matching EfficientNet-B7.
Unified Latent Dynamics unifies model-free and model-based reinforcement learning.
problem Combining the efficiency of model-free methods with the representational strengths of model-based approaches.
method Embedding state-action pairs into a latent space where the true value function is approximately linear, using synchronized updates of encoder, value, and policy networks.
result ULD achieves cross-domain competence with minimal tuning and a fraction of the parameter footprint.
Continuous-time MBRL framework tackles control systems with Bayesian ODEs.
problem Discretization of continuous-time systems in MBRL.
method Novel actor-critic method with Bayesian ODEs for state inference.
result Model robust against irregular and noisy data, sample-efficient, solves challenging control problems.
Our work proves CSF can recover ground-truth features in RL, improving understanding of feature learning.
problem Understanding the role of representation and mutual information in reinforcement learning.
method Investigates Contrastive Successor Features (CSF) method for identifiable representation learning in reinforcement learning.
result Proves CSF can recover ground-truth features up to a linear transformation.
Continuous control tasks in reinforcement learning are important because they provide an important framework for learning in high-dimensional state spaces with deceptive rewards, where the agent can easily become trapped into suboptimal solutions. One way to avoid local optima is to use a population of agents to ensure…
New diffusion models improve counterfactual image generation with semantic control.
problem Challenges in preserving identity, maintaining quality, and ensuring causal model faithfulness in counterfactual image generation.
method Integrates semantic representations into diffusion models through Pearlian causality, introducing spatial, semantic, and dynamic abduction.
result Demonstrates high-level semantic identity preservation and principled trade-offs between faithful causal control and identity preservation.
A new framework for controllable generation of discrete masked models.
problem Efficient controllable generation of discrete data models.
method Plug-and-play framework based on importance sampling.
result Demonstrates versatility across multiple domains, including protein design.
DeepFDR uses deep learning for better FDR control in neuroimaging data.
problem Spatial dependence among voxel-based tests in neuroimaging data.
method DeepFDR leverages unsupervised deep learning-based image segmentation.
result DeepFDR outperforms existing methods in FDR control and computational efficiency.
Neural nets improve plasma equilibrium modeling for NSTX-U.
problem Modeling plasma equilibrium and shape control for NSTX-U.
method Developed two neural networks: Eqnet and Pertnet.
result NNs offer faster and more flexible prediction of plasma scenarios.