Transformers learn sparse Boolean functions through RL and SFT, revealing distinct learning behaviors.
problem Learning sparse Boolean functions with Transformers.
method Reinforcement Learning (RL) with process rewards and Supervised Fine-Tuning (SFT).
result RL learns the whole CoT chain simultaneously, while SFT learns step by step.
Equivariant CNNs improve RL performance in symmetric environments.
problem Learning equivariant representations for RL in symmetric environments.
method Proposed and studied equivariant CNNs for RL.
result Equivariant CNNs enhance RL performance and sample efficiency.
Pre-trained LLM adapted with LoRA improves offline RL for quantitative trading.
problem Challenges in offline RL for quantitative trading due to complex temporal dependencies and overfitting.
method Integrates pre-trained GPT-2 weights and LoRA for efficient fine-tuning of a Decision Transformer.
result Outperforms existing offline RL methods in certain trading scenarios.
Study reveals RL game's embedding space is stratified, not a manifold.
problem Understanding the structure of RL game embeddings.
method Adapted Robinson's volume growth transform for RL setting.
result Token embedding space is stratified, not a manifold.
Transformer RL optimizes A/B testing for time series experiments.
problem Challenges in applying A/B testing to time series experiments, especially with limited history and strong assumptions.
method Transformer reinforcement learning approach that conditions allocation on full history and optimizes MSE without restrictive assumptions.
result Consistently outperforms existing designs in synthetic, simulator, and real-world data.
This paper shows using classification instead of regression improves deep RL scalability.
problem Challenges in training value functions for large networks in deep RL.
method Used categorical cross-entropy loss instead of mean squared error regression.
result Significant improvements in performance and scalability across various domains.
Owing to their ability to both effectively integrate information over long time horizons and scale to massive amounts of data, self-attention architectures have recently shown breakthrough success in natural language processing (NLP), achieving state-of-the-art results in domains such as language modeling and machine t…
New RL method learns from state transitions without actions.
problem Offline RL with missing action labels.
method State policy discretisation and decQN algorithm.
result Improves convergence speed and performance in online RL.
ICEE learns new RL tasks in less time with a Transformer model.
problem Efficient in-context policy learning for reinforcement learning.
method In-context Exploration-Exploitation (ICEE) algorithm that optimizes efficiency without explicit Bayesian inference.
result ICEE solves Bayesian optimization problems as efficiently as Gaussian process biased methods but in significantly less time.
SUPE combines unlabeled data with RL to efficiently explore tasks.
problem Efficient exploration in reinforcement learning with sparse rewards.
method Extract low-level skills using VAE, pseudo-label unlabeled data, and use as off-policy data for online RL.
result SUPE outperforms prior methods across 42 long-horizon tasks.
Theoretical analysis of data quality and synergies in LLMs.
problem Understanding why different training methods require different amounts of data.
method Theoretical analysis of transformers trained on a weight prediction task for linear regression.
result SFT excels on smaller datasets challenging for the pretrained model, while RL benefits from large, not overly difficult data.
Transforming sparse outcomes into dense process rewards for efficient reinforcement learning.
problem Training RL policies to maximize sparse outcomes.
method Incentivizing policy matching state-action visitations of successful episodes.
result Significantly faster RL finetuning performance.
Study proposes adaptive RL for dynamic portfolio optimization.
problem Traditional portfolio optimization models fail to adapt to regime shifts.
method Regime-aware reinforcement learning framework with hybrid observations and constrained reward functions.
result Transformer PPO achieves highest risk-adjusted returns, while LSTM variants offer a good balance.
The lottery ticket hypothesis proposes that over-parameterization of deep neural networks (DNNs) aids training by increasing the probability of a "lucky" sub-network initialization being present rather than by helping the optimization process (Frankle & Carbin, 2019). Intriguingly, this phenomenon suggests that initial…
A new RL model ensures safe learning in uncertain environments.
problem Safe reinforcement learning in uncertain, partially observable environments.
method Lyapunov-based uncertainty quantification and Transformers for memory.
result Significant improvement in safety and optimality in grid-world tasks.
In this paper we present Horizon, Facebook's open source applied reinforcement learning (RL) platform. Horizon is an end-to-end platform designed to solve industry applied RL problems where datasets are large (millions to billions of observations), the feedback loop is slow (vs. a simulator), and experiments must be do…
Sentiment analysis from LLMs improves financial trading performance.
problem Improving dynamic strategy optimization in financial markets.
method Integration of sentiment analysis from LLMs into RL frameworks.
result Sentiment-enhanced RL models outperform traditional RL models in net worth and cumulative profit.
Improved risk-sensitive RL with exponential Bellman equation and better regret bounds.
problem Exponential gap between upper and lower bounds in risk-sensitive RL.
method Identified and addressed deficiencies in existing algorithms and analysis; developed novel analysis and exploration mechanism.
result Improved regret upper bounds over existing ones.
Transformer learns to search through reinforcement learning, mimicking DFS.
problem Understanding how transformers learn search capabilities in RL.
method Two-head transformer, depth-wise curriculum, discounted returns.
result Transformer policy generalizes depth and prioritizes high-probability branches.
Supervised learning alone can be effective for offline RL, revealing essential elements.
problem Understanding when and how supervised learning alone can be effective for offline RL.
method Extensive experiments to identify essential elements for offline RL via supervised learning.
result Maximizing likelihood with a two-layer feedforward MLP is competitive with more complex methods.
Enhances RL in target domains with limited data using augmented return.
problem Utilize data from an accessible source domain to improve policy learning in a target domain with scarce data.
method Return Augmented Decision Transformer (REAG) method, which augments the return in the source domain to align with the target domain's optimal trajectory distribution.
result The proposed REAG method achieves the same level of suboptimality as without a dynamics shift, enhancing DT type frameworks' performance in off-dynamics RL.
We present the Compressive Transformer, an attentive sequence model which compresses past memories for long-range sequence learning. We find the Compressive Transformer obtains state-of-the-art language modelling results in the WikiText-103 and Enwik8 benchmarks, achieving 17.1 ppl and 0.97 bpc respectively. We also fi…
Deep RL controls anesthesia more accurately than traditional methods.
problem Controlling the level of unconsciousness during anesthesia.
method Deep Reinforcement Learning (DRL) to map patient state to propofol dosage.
result Deep RL model outperformed traditional controllers (1.7% vs 3.4% median absolute performance error).
GDT improves reinforcement learning by matching future state information efficiently.
problem Efficient learning of multi-task policies from trajectory data.
method Generalized Decision Transformer (GDT) for offline hindsight information matching.
result GDT enables effective offline multi-task state-marginal matching and imitation learning.
RECODE uses clustering and embedding to track state visitation counts in RL.
problem Efficient novelty-based exploration in nonstationary RL environments.
method Non-parametric clustering, online density estimation, inverse dynamics loss.
result RECODE achieves state-of-the-art performance in challenging RL tasks.
AlphaGrad optimizes memory usage in RL algorithms by normalizing gradients.
problem Memory overhead and hyperparameter complexity in adaptive optimizers.
method Tensor-wise L2 normalization followed by a smooth hyperbolic tangent transformation controlled by a single parameter.
result AlphaGrad provides enhanced training stability and competitive performance in various RL algorithms.
QAM uses adjoint matching to optimize continuous-action RL policies efficiently.
problem Efficient optimization of expressive diffusion or flow-matching policies with respect to a Q-function.
method QAM leverages adjoint matching to bypass the numerical instability of backpropagation through multi-step denoising processes.
result QAM consistently outperforms prior approaches on hard, sparse reward tasks in offline and offline-to-online RL.
Transformers learn to play games in-context, proving Nash equilibrium.
problem Understanding in-context game-playing capabilities of pre-trained transformers.
method Theoretical guarantees and constructional results for transformer architecture in multi-agent games.
result Pre-trained transformers can learn Nash equilibrium in-context for two-player zero-sum games.
Transformers can implement reinforcement learning algorithms from data without updates.
problem Training reinforcement learning algorithms from data without parameter updates.
method Design a teacher-mimicking training procedure for transformers to implement policy-improvement methods.
result Gradient flow converges to an optimal parameter manifold corresponding to the desired RL update.
The paper explores how duality applies to reinforcement learning.
problem Exploring the use of duality in reinforcement learning.
method Demonstrates duality in various RL methods, including value iteration and approximations.
result Lagrangian duality is found to be prevalent in reinforcement learning.
Open-domain dialog generation is a challenging problem; maximum likelihood training can lead to repetitive outputs, models have difficulty tracking long-term conversational goals, and training on standard movie or online datasets may lead to the generation of inappropriate, biased, or offensive text. Reinforcement Lear…
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.
New RL method learns to skip states in linearly qπ-realizable MDPs, simplifying to linear MDPs.
problem Online RL in episodic MDPs with linearly qπ-realizable action-values. method Derives a novel algorithm that learns to skip states and applies a linear MDP algorithm.
result First polynomial-sample-complexity online RL algorithm for linearly qπ-realizable MDPs. Optimizes reinforcement learning by prioritizing sets of samples over individual ones.
problem Limits exploration and improvement on harder examples due to focusing on isolated samples.
method Proposes Pass@K Policy Optimization (PKPO) to optimize for sets of samples that maximize reward when considered jointly.
result Optimization with novel low variance unbiased estimators for pass@k and its gradient leads to significant pass@k gains.
Manipulating data, such as weighting data examples or augmenting with new instances, has been increasingly used to improve model training. Previous work has studied various rule- or learning-based approaches designed for specific types of data manipulation. In this work, we propose a new method that supports learning d…
Study optimal stopping for diffusion processes with unknown primitives, applying RL and martingale methods.
problem Optimal stopping for diffusion processes with unknown model primitives.
method Continuous-time reinforcement learning framework, variational inequality formulation, stochastic optimal control, entropy regularizer, semi-analytical optimal Bernoulli distribution, policy improvement theorem, policy iterations.
result Demonstrated high accuracy in learning value functions and characterizing free boundaries for various optimal stopping problems.
In many real-world reinforcement learning (RL) problems, besides optimizing the main objective function, an agent must concurrently avoid violating a number of constraints. In particular, besides optimizing performance it is crucial to guarantee the safety of an agent during training as well as deployment (e.g. a robot…
The H∞ control design problem is considered for nonlinear systems with unknown internal system model. It is known that the nonlinear H∞ control problem can be transformed into solving the so-called Hamilton-Jacobi-Isaacs (HJI) equation, which is a nonlinear partial differential equation that is genera…
Study uses RL to optimize global equity portfolios, finds mixed results.
problem Optimizing dynamic portfolio weights across diverse global markets.
method Deep reinforcement learning with Soft Actor-Critic, incorporating various constraints and reward formulations.
result RL strategies achieve competitive performance, but no strategy consistently outperforms Buy and Hold.
The UNKNOT problem solved using natural language processing and machine learning.
problem Determining if a knot is the unknot.
method Braid word representation, binary classification, Reformer and shared-QK Transformer networks, reinforcement learning, Markov moves, braid relations.
result Reformer and shared-QK Transformer networks outperform fully-connected networks in predicting the unknot.
Hybrid RL algorithms improve offline and online RL in linear MDPs.
problem Improving RL performance without single-policy concentrability.
method Developed computationally efficient algorithms for PAC and regret-minimizing RL in linear MDPs.
result Achieved sharper error or regret bounds for linear MDPs.
This project proposes using reinforcement learning to train spiking neural networks.
problem Training spiking neural networks using traditional methods is challenging due to the discrete nature of spikes.
method The project investigates two approaches: 1) treating each neuron as an RL agent, 2) applying the reparameterization trick.
result The project demonstrates that reinforcement learning can be applied to train spiking neural networks.
This work explores representation complexity in RL paradigms, revealing model-based RL as the easiest task.
problem Investigating the representation complexity gap among model-based, policy-based, and value-based RL.
method Demonstrated through analysis of Markov decision processes (MDPs) and introduced new classes of MDPs.
result Representation complexity hierarchy: model-based RL > policy-based RL > value-based RL.
New method defends RL agents from poisoning attacks without MDP knowledge.
problem Poisoning attacks on RL systems can cause learning failures.
method Generic poisoning framework for online RL, Vulnerability-Aware Adversarial Critic Poison (VA2C-P).
result Successfully prevents RL agents from learning good policies or converging to target policies.
With the advent of deep generative models in computational chemistry, in silico anticancer drug design has undergone an unprecedented transformation. While state-of-the-art deep learning approaches have shown potential in generating compounds with desired chemical properties, they disregard the genetic profile and prop…
When performing imitation learning from expert demonstrations, distribution matching is a popular approach, in which one alternates between estimating distribution ratios and then using these ratios as rewards in a standard reinforcement learning (RL) algorithm. Traditionally, estimation of the distribution ratio requi…
Reincarnating RL reuses prior work to accelerate RL progress.
problem Efficiency and accessibility in reinforcement learning for large-scale applications.
method Transfer of learned policies between RL agents or design iterations, focusing on value-based RL.
result Demonstrated gains in performance over tabula rasa RL on various tasks.
Paper shows RLHF can be solved similarly to standard RL.
problem Difficulty of RLHF compared to standard RL.
method Reduction to reward-based RL techniques.
result RLHF can be solved using existing algorithms for reward-based RL.