Search

Offline-to-Online Reinforcement Learning has emerged as a powerful paradigm, leveraging offline data for initialization and online fine-tuning to enhance both sample efficiency and performance. However, most existing research has focused on single-agent settings, with limited exploration of the multi-agent extension, i.e., Offline-to-Online Multi-Agent Reinforcement Learning (O2O MARL). In O2O MARL, two critical challenges become more prominent as the number of agents increases: (i) the risk of unlearning pre-trained Q-values due to distributional shifts during the transition from offline-to-online phases, and (ii) the difficulty of efficient exploration in the large joint state-action space. To tackle these challenges, we propose a novel O2O MARL framework called Offline Value Function Memory with Sequential Exploration (OVMSE). First, we introduce the Offline Value Function Memory (OVM) mechanism to compute target Q-values, preserving knowledge gained during offline training, ensuring smoother transitions, and enabling efficient fine-tuning. Second, we propose a decentralized Sequential Exploration (SE) strategy tailored for O2O MARL, which effectively utilizes the pre-trained offline policy for exploration, thereby significantly reducing the joint state-action space to be explored. Extensive experiments on the StarCraft Multi-Agent Challenge (SMAC) demonstrate that OVMSE significantly outperforms existing baselines, achieving superior sample efficiency and overall performance.

In this paper, we present a novel algorithm, uniINF, for the Heavy-Tailed Multi-Armed Bandits (HTMAB) problem, demonstrating robustness and adaptability in both stochastic and adversarial environments. Unlike the stochastic MAB setting where loss distributions are stationary with time, our study extends to the adversarial setup, where losses are generated from heavy-tailed distributions that depend on both arms and time. Our novel algorithm `uniINF` enjoys the so-called Best-of-Both-Worlds (BoBW) property, performing optimally in both stochastic and adversarial environments without knowing the exact environment type. Moreover, our algorithm also possesses a Parameter-Free feature, i.e., it operates without the need of knowing the heavy-tail parameters $(\sigma, \alpha)$ a-priori. To be precise, uniINF ensures nearly-optimal regret in both stochastic and adversarial environments, matching the corresponding lower bounds when $(\sigma, \alpha)$ is known (up to logarithmic factors). To our knowledge, uniINF is the first parameter-free algorithm to achieve the BoBW property for the heavy-tailed MAB problem. Technically, we develop innovative techniques to achieve BoBW guarantees for Parameter-Free HTMABs, including a refined analysis for the dynamics of log-barrier, an auto-balancing learning rate scheduling scheme, an adaptive skipping-clipping loss tuning technique, and a stopping-time analysis for logarithmic regret.

Generative Flow Networks (GFlowNets) are a novel class of generative models designed to sample from unnormalized distributions and have found applications in various important tasks, attracting great research interest in their training algorithms. In general, GFlowNets are trained by fitting the forward flow to the backward flow on sampled training objects. Prior work focused on the choice of training objects, parameterizations, sampling and resampling strategies, and backward policies, aiming to enhance credit assignment, exploration, or exploitation of the training process. However, the choice of regression loss, which can highly influence the exploration and exploitation behavior of the under-training policy, has been overlooked. Due to the lack of theoretical understanding for choosing an appropriate regression loss, most existing algorithms train the flow network by minimizing the squared error of the forward and backward flows in log-space, i.e., using the quadratic regression loss. In this work, we rigorously prove that distinct regression losses correspond to specific divergence measures, enabling us to design and analyze regression losses according to the desired properties of the corresponding divergence measures. Specifically, we examine two key properties: zero-forcing and zero-avoiding, where the former promotes exploitation and higher rewards, and the latter encourages exploration and enhances diversity. Based on our theoretical framework, we propose three novel regression losses, namely, Shifted-Cosh, Linex(1/2), and Linex(1). We evaluate them across three benchmarks: hyper-grid, bit-sequence generation, and molecule generation. Our proposed losses are compatible with most existing training algorithms, and significantly improve the performances of the algorithms concerning convergence speed, sample diversity, and robustness.

Deep Multi-agent Reinforcement Learning (MARL) relies on neural networks with numerous parameters in multi-agent scenarios, often incurring substantial computational overhead. Consequently, there is an urgent need to expedite training and enable model compression in MARL. This paper proposes the utilization of dynamic sparse training (DST), a technique proven effective in deep supervised learning tasks, to alleviate the computational burdens in MARL training. However, a direct adoption of DST fails to yield satisfactory MARL agents, leading to breakdowns in value learning within deep sparse value-based MARL models. Motivated by this challenge, we introduce an innovative Multi-Agent Sparse Training (MAST) framework aimed at simultaneously enhancing the reliability of learning targets and the rationality of sample distribution to improve value learning in sparse models. Specifically, MAST incorporates the Soft Mellowmax Operator with a hybrid TD-($\lambda$) schema to establish dependable learning targets. Additionally, it employs a dual replay buffer mechanism to enhance the distribution of training samples. Building upon these aspects, MAST utilizes gradient-based topology evolution to exclusively train multiple MARL agents using sparse networks. Our comprehensive experimental investigation across various value-based MARL algorithms on multiple benchmarks demonstrates, for the first time, significant reductions in redundancy of up to $20\times$ in Floating Point Operations (FLOPs) for both training and inference, with less than $3\%$ performance degradation.

Stochastic Network Optimization (SNO) concerns scheduling in stochastic queueing systems. It has been widely studied in network theory. Classical SNO algorithms require network conditions to be stationary with time, which fails to capture the non-stationary components in many real-world scenarios. Many existing algorithms also assume knowledge of network conditions before decision, which rules out applications where unpredictability presents. Motivated by these issues, we consider Adversarial Network Optimization (ANO) under bandit feedback. Specifically, we consider the task of *i)* maximizing some unknown and time-varying utility function associated to scheduler's actions, where *ii)* the underlying network is a non-stationary multi-hop one whose conditions change arbitrarily with time, and *iii)* only bandit feedback (effect of actually deployed actions) is revealed after decisions. Our proposed `UMO2` algorithm ensures network stability and also matches the utility maximization performance of any "mildly varying" reference policy up to a polynomially decaying gap. To our knowledge, no previous ANO algorithm handled multi-hop networks or achieved utility guarantees under bandit feedback, whereas ours can do both. Technically, our method builds upon a novel integration of online learning into Lyapunov analyses: To handle complex inter-dependencies among queues in multi-hop networks, we propose meticulous techniques to balance online learning and Lyapunov arguments. To tackle the learning obstacles due to potentially unbounded queue sizes, we design a new online linear optimization algorithm that automatically adapts to loss magnitudes. To maximize utility, we propose a bandit convex optimization algorithm with novel queue-dependent learning rate scheduling that suites drastically varying queue lengths. Our new insights in online learning can be of independent interest.

Large language models (LLMs) have made significant strides in complex tasks, yet their widespread adoption is impeded by substantial computational demands. With hundreds of billion parameters, transformer-based LLMs necessitate months of pretraining across a high-end GPU cluster. However, this paper reveals a compelling finding: transformers exhibit considerable redundancy in pretraining computations, which motivates our proposed solution, Mixed Sparsity Training (MST), an efficient pretraining method that can reduce about $75\%$ of Floating Point Operations (FLOPs) while maintaining performance. MST integrates dynamic sparse training (DST) with Sparsity Variation (SV) and Hybrid Sparse Attention (HSA) during pretraining, involving three distinct phases: warm-up, ultra-sparsification, and restoration. The warm-up phase transforms the dense model into a sparse one, and the restoration phase reinstates connections. Throughout these phases, the model is trained with a dynamically evolving sparse topology and an HSA mechanism to maintain performance and minimize training FLOPs concurrently. Our experiment on GPT-2 showcases a FLOP reduction of $4\times$ without compromising performance.

Effective action abstraction is crucial in tackling challenges associated with large action spaces in Imperfect Information Extensive-Form Games (IIEFGs). However, due to the vast state space and computational complexity in IIEFGs, existing methods often rely on fixed abstractions, resulting in sub-optimal performance. In response, we introduce RL-CFR, a novel reinforcement learning (RL) approach for dynamic action abstraction. RL-CFR builds upon our innovative Markov Decision Process (MDP) formulation, with states corresponding to public information and actions represented as feature vectors indicating specific action abstractions. The reward is defined as the expected payoff difference between the selected and default action abstractions. RL-CFR constructs a game tree with RL-guided action abstractions and utilizes counterfactual regret minimization (CFR) for strategy derivation. Impressively, it can be trained from scratch, achieving higher expected payoff without increased CFR solving time. In experiments on Heads-up No-limit Texas Hold'em, RL-CFR outperforms ReBeL's replication and Slumbot, demonstrating significant win-rate margins of $64\pm 11$ and $84\pm 17$ mbb/hand, respectively.

In the realm of reinforcement learning (RL), accounting for risk is crucial for making decisions under uncertainty, particularly in applications where safety and reliability are paramount. In this paper, we introduce a general framework on Risk-Sensitive Distributional Reinforcement Learning (RS-DisRL), with static Lipschitz Risk Measures (LRM) and general function approximation. Our framework covers a broad class of risk-sensitive RL, and facilitates analysis of the impact of estimation functions on the effectiveness of RSRL strategies and evaluation of their sample complexity. We design two innovative meta-algorithms: \texttt{RS-DisRL-M}, a model-based strategy for model-based function approximation, and \texttt{RS-DisRL-V}, a model-free approach for general value function approximation. With our novel estimation techniques via Least Squares Regression (LSR) and Maximum Likelihood Estimation (MLE) in distributional RL with augmented Markov Decision Process (MDP), we derive the first $\widetilde{\mathcal{O}}(\sqrt{K})$ dependency of the regret upper bound for RSRL with static LRM, marking a pioneering contribution towards statistically efficient algorithms in this domain.

This work pioneers regret analysis of risk-sensitive reinforcement learning in partially observable environments with hindsight observation, addressing a gap in theoretical exploration. We introduce a novel formulation that integrates hindsight observations into a Partially Observable Markov Decision Process (POMDP) framework, where the goal is to optimize accumulated reward under the entropic risk measure. We develop the first provably efficient RL algorithm tailored for this setting. We also prove by rigorous analysis that our algorithm achieves polynomial regret $\tilde{O}\left(\frac{e^{|{\gamma}|H}-1}{|{\gamma}|H}H^2\sqrt{KHS^2OA}\right)$, which outperforms or matches existing upper bounds when the model degenerates to risk-neutral or fully observable settings. We adopt the method of change-of-measure and develop a novel analytical tool of beta vectors to streamline mathematical derivations. These techniques are of particular interest to the theoretical study of reinforcement learning., Comment: 38 pages

Latent Consistency Models (LCMs) have achieved impressive performance in accelerating text-to-image generative tasks, producing high-quality images with minimal inference steps. LCMs are distilled from pre-trained latent diffusion models (LDMs), requiring only ~32 A100 GPU training hours. This report further extends LCMs' potential in two aspects: First, by applying LoRA distillation to Stable-Diffusion models including SD-V1.5, SSD-1B, and SDXL, we have expanded LCM's scope to larger models with significantly less memory consumption, achieving superior image generation quality. Second, we identify the LoRA parameters obtained through LCM distillation as a universal Stable-Diffusion acceleration module, named LCM-LoRA. LCM-LoRA can be directly plugged into various Stable-Diffusion fine-tuned models or LoRAs without training, thus representing a universally applicable accelerator for diverse image generation tasks. Compared with previous numerical PF-ODE solvers such as DDIM, DPM-Solver, LCM-LoRA can be viewed as a plug-in neural PF-ODE solver that possesses strong generalization abilities. Project page: https://github.com/luosiallen/latent-consistency-model., Comment: Technical Report

In distributed deep learning with data parallelism, synchronizing gradients at each training step can cause a huge communication overhead, especially when many nodes work together to train large models. Local gradient methods, such as Local SGD, address this issue by allowing workers to compute locally for $H$ steps without synchronizing with others, hence reducing communication frequency. While $H$ has been viewed as a hyperparameter to trade optimization efficiency for communication cost, recent research indicates that setting a proper $H$ value can lead to generalization improvement. Yet, selecting a proper $H$ is elusive. This work proposes a theory-grounded method for determining $H$, named the Quadratic Synchronization Rule (QSR), which recommends dynamically setting $H$ in proportion to $\frac{1}{\eta^2}$ as the learning rate $\eta$ decays over time. Extensive ImageNet experiments on ResNet and ViT show that local gradient methods with QSR consistently improve the test accuracy over other synchronization strategies. Compared with the standard data parallel training, QSR enables Local AdamW on ViT-B to cut the training time on 16 or 64 GPUs down from 26.7 to 20.2 hours or from 8.6 to 5.5 hours and, at the same time, achieves $1.16\%$ or $0.84\%$ higher top-1 validation accuracy., Comment: camera-ready version for ICLR'24

Deep reinforcement learning has achieved remarkable performance in various domains by leveraging deep neural networks for approximating value functions and policies. However, using neural networks to approximate value functions or policy functions still faces challenges, including low sample efficiency and overfitting. In this paper, we introduce OMNet, a novel learning paradigm utilizing multiple subnetworks within a single network, offering diverse outputs efficiently. We provide a systematic pipeline, including initialization, training, and sampling with OMNet. OMNet can be easily applied to various deep reinforcement learning algorithms with minimal additional overhead. Through comprehensive evaluations conducted on MuJoCo benchmark, our findings highlight OMNet's ability to strike an effective balance between performance and computational cost., Comment: Preprint

Latent Diffusion models (LDMs) have achieved remarkable results in synthesizing high-resolution images. However, the iterative sampling process is computationally intensive and leads to slow generation. Inspired by Consistency Models (song et al.), we propose Latent Consistency Models (LCMs), enabling swift inference with minimal steps on any pre-trained LDMs, including Stable Diffusion (rombach et al). Viewing the guided reverse diffusion process as solving an augmented probability flow ODE (PF-ODE), LCMs are designed to directly predict the solution of such ODE in latent space, mitigating the need for numerous iterations and allowing rapid, high-fidelity sampling. Efficiently distilled from pre-trained classifier-free guided diffusion models, a high-quality 768 x 768 2~4-step LCM takes only 32 A100 GPU hours for training. Furthermore, we introduce Latent Consistency Fine-tuning (LCF), a novel method that is tailored for fine-tuning LCMs on customized image datasets. Evaluation on the LAION-5B-Aesthetics dataset demonstrates that LCMs achieve state-of-the-art text-to-image generation performance with few-step inference. Project Page: https://latent-consistency-models.github.io/

Risk-sensitive reinforcement learning (RL) aims to optimize policies that balance the expected reward and risk. In this paper, we present a novel risk-sensitive RL framework that employs an Iterated Conditional Value-at-Risk (CVaR) objective under both linear and general function approximations, enriched by human feedback. These new formulations provide a principled way to guarantee safety in each decision making step throughout the control process. Moreover, integrating human feedback into risk-sensitive RL framework bridges the gap between algorithmic decision-making and human participation, allowing us to also guarantee safety for human-in-the-loop systems. We propose provably sample-efficient algorithms for this Iterated CVaR RL and provide rigorous theoretical analysis. Furthermore, we establish a matching lower bound to corroborate the optimality of our algorithms in a linear context.

We present a novel Diffusion Offline Multi-agent Model (DOM2) for offline Multi-Agent Reinforcement Learning (MARL). Different from existing algorithms that rely mainly on conservatism in policy design, DOM2 enhances policy expressiveness and diversity based on diffusion. Specifically, we incorporate a diffusion model into the policy network and propose a trajectory-based data-augmentation scheme in training. These key ingredients make our algorithm more robust to environment changes and achieve significant improvements in performance, generalization and data-efficiency. Our extensive experimental results demonstrate that DOM2 outperforms existing state-of-the-art methods in multi-agent particle and multi-agent MuJoCo environments, and generalizes significantly better in shifted environments thanks to its high expressiveness and diversity. Furthermore, DOM2 shows superior data efficiency and can achieve state-of-the-art performance with $20+$ times less data compared to existing algorithms.

In this paper, we study scheduling of a queueing system with zero knowledge of instantaneous network conditions. We consider a one-hop single-server queueing system consisting of $K$ queues, each with time-varying and non-stationary arrival and service rates. Our scheduling approach builds on an innovative combination of adversarial bandit learning and Lyapunov drift minimization, without knowledge of the instantaneous network state (the arrival and service rates) of each queue. We then present two novel algorithms \texttt{SoftMW} (SoftMaxWeight) and \texttt{SSMW} (Sliding-window SoftMaxWeight), both capable of stabilizing systems that can be stablized by some (possibly unknown) sequence of randomized policies whose time-variation satisfies a mild condition. We further generalize our results to the setting where arrivals and departures only have bounded moments instead of being deterministically bounded and propose \texttt{SoftMW+} and \texttt{SSMW+} that are capable of stabilizing the system. As a building block of our new algorithms, we also extend the classical \texttt{EXP3.S} (Auer et al., 2002) algorithm for multi-armed bandits to handle unboundedly large feedback signals, which can be of independent interest.

Local SGD is a communication-efficient variant of SGD for large-scale training, where multiple GPUs perform SGD independently and average the model parameters periodically. It has been recently observed that Local SGD can not only achieve the design goal of reducing the communication overhead but also lead to higher test accuracy than the corresponding SGD baseline (Lin et al., 2020b), though the training regimes for this to happen are still in debate (Ortiz et al., 2021). This paper aims to understand why (and when) Local SGD generalizes better based on Stochastic Differential Equation (SDE) approximation. The main contributions of this paper include (i) the derivation of an SDE that captures the long-term behavior of Local SGD in the small learning rate regime, showing how noise drives the iterate to drift and diffuse after it has reached close to the manifold of local minima, (ii) a comparison between the SDEs of Local SGD and SGD, showing that Local SGD induces a stronger drift term that can result in a stronger effect of regularization, e.g., a faster reduction of sharpness, and (iii) empirical evidence validating that having a small learning rate and long enough training time enables the generalization improvement over SGD but removing either of the two conditions leads to no improvement., Comment: Published as a conference paper at ICLR 2023

Inspired by the recent success of sequence modeling in RL and the use of masked language model for pre-training, we propose a masked model for pre-training in RL, RePreM (Representation Pre-training with Masked Model), which trains the encoder combined with transformer blocks to predict the masked states or actions in a trajectory. RePreM is simple but effective compared to existing representation pre-training methods in RL. It avoids algorithmic sophistication (such as data augmentation or estimating multiple models) with sequence modeling and generates a representation that captures long-term dynamics well. Empirically, we demonstrate the effectiveness of RePreM in various tasks, including dynamic prediction, transfer learning, and sample-efficient RL with both value-based and actor-critic methods. Moreover, we show that RePreM scales well with dataset size, dataset quality, and the scale of the encoder, which indicates its potential towards big RL models., Comment: Accepted by AAAI-23

Generative Flow Networks (or GFlowNets for short) are a family of probabilistic agents that learn to sample complex combinatorial structures through the lens of "inference as control". They have shown great potential in generating high-quality and diverse candidates from a given energy landscape. However, existing GFlowNets can be applied only to deterministic environments, and fail in more general tasks with stochastic dynamics, which can limit their applicability. To overcome this challenge, this paper introduces Stochastic GFlowNets, a new algorithm that extends GFlowNets to stochastic environments. By decomposing state transitions into two steps, Stochastic GFlowNets isolate environmental stochasticity and learn a dynamics model to capture it. Extensive experimental results demonstrate that Stochastic GFlowNets offer significant advantages over standard GFlowNets as well as MCMC- and RL-based approaches, on a variety of standard benchmarks with stochastic dynamics., Comment: UAI 2023

In this paper, we investigate a novel safe reinforcement learning problem with step-wise violation constraints. Our problem differs from existing works in that we consider stricter step-wise violation constraints and do not assume the existence of safe actions, making our formulation more suitable for safety-critical applications which need to ensure safety in all decision steps and may not always possess safe actions, e.g., robot control and autonomous driving. We propose a novel algorithm SUCBVI, which guarantees $\widetilde{O}(\sqrt{ST})$ step-wise violation and $\widetilde{O}(\sqrt{H^3SAT})$ regret. Lower bounds are provided to validate the optimality in both violation and regret performance with respect to $S$ and $T$. Moreover, we further study a novel safe reward-free exploration problem with step-wise violation constraints. For this problem, we design an $(\varepsilon,\delta)$-PAC algorithm SRF-UCRL, which achieves nearly state-of-the-art sample complexity $\widetilde{O}((\frac{S^2AH^2}{\varepsilon}+\frac{H^4SA}{\varepsilon^2})(\log(\frac{1}{\delta})+S))$, and guarantees $\widetilde{O}(\sqrt{ST})$ violation during the exploration. The experimental results demonstrate the superiority of our algorithms in safety performance, and corroborate our theoretical results.

Despite the recent success of representation learning in sequential decision making, the study of the pure exploration scenario (i.e., identify the best option and minimize the sample complexity) is still limited. In this paper, we study multi-task representation learning for best arm identification in linear bandits (RepBAI-LB) and best policy identification in contextual linear bandits (RepBPI-CLB), two popular pure exploration settings with wide applications, e.g., clinical trials and web content optimization. In these two problems, all tasks share a common low-dimensional linear representation, and our goal is to leverage this feature to accelerate the best arm (policy) identification process for all tasks. For these problems, we design computationally and sample efficient algorithms DouExpDes and C-DouExpDes, which perform double experimental designs to plan optimal sample allocations for learning the global representation. We show that by learning the common representation among tasks, our sample complexity is significantly better than that of the native approach which solves tasks independently. To the best of our knowledge, this is the first work to demonstrate the benefits of representation learning for multi-task pure exploration.

We propose Banker Online Mirror Descent (Banker-OMD), a novel framework generalizing the classical Online Mirror Descent (OMD) technique in the online learning literature. The Banker-OMD framework almost completely decouples feedback delay handling and the task-specific OMD algorithm design, thus facilitating the design of new algorithms capable of efficiently and robustly handling feedback delays. Specifically, it offers a general methodology for achieving $\widetilde{\mathcal O}(\sqrt{T} + \sqrt{D})$-style regret bounds in online bandit learning tasks with delayed feedback, where $T$ is the number of rounds and $D$ is the total feedback delay. We demonstrate the power of \texttt{Banker-OMD} by applications to two important bandit learning scenarios with delayed feedback, including delayed scale-free adversarial Multi-Armed Bandits (MAB) and delayed adversarial linear bandits. \texttt{Banker-OMD} leads to the first delayed scale-free adversarial MAB algorithm achieving $\widetilde{\mathcal O}(\sqrt{K}L(\sqrt T+\sqrt D))$ regret and the first delayed adversarial linear bandit algorithm achieving $\widetilde{\mathcal O}(\text{poly}(n)(\sqrt{T} + \sqrt{D}))$ regret. As a corollary, the first application also implies $\widetilde{\mathcal O}(\sqrt{KT}L)$ regret for non-delayed scale-free adversarial MABs, which is the first to match the $\Omega(\sqrt{KT}L)$ lower bound up to logarithmic factors and can be of independent interest., Comment: Based on preliminary works arXiv:2106.08943 and arXiv:2110.13400; Accepted by ICML2023

We study a general multi-dueling bandit problem, where an agent compares multiple options simultaneously and aims to minimize the regret due to selecting suboptimal arms. This setting generalizes the traditional two-dueling bandit problem and finds many real-world applications involving subjective feedback on multiple options. We start with the two-dueling bandit setting and propose two efficient algorithms, DoublerBAI and MultiSBM-Feedback. DoublerBAI provides a generic schema for translating known results on best arm identification algorithms to the dueling bandit problem, and achieves a regret bound of $O(\ln T)$. MultiSBM-Feedback not only has an optimal $O(\ln T)$ regret, but also reduces the constant factor by almost a half compared to benchmark results. Then, we consider the general multi-dueling case and develop an efficient algorithm MultiRUCB. Using a novel finite-time regret analysis for the general multi-dueling bandit problem, we show that MultiRUCB also achieves an $O(\ln T)$ regret bound and the bound tightens as the capacity of the comparison set increases. Based on both synthetic and real-world datasets, we empirically demonstrate that our algorithms outperform existing algorithms.

The Generative Flow Network is a probabilistic framework where an agent learns a stochastic policy for object generation, such that the probability of generating an object is proportional to a given reward function. Its effectiveness has been shown in discovering high-quality and diverse solutions, compared to reward-maximizing reinforcement learning-based methods. Nonetheless, GFlowNets only learn from rewards of the terminal states, which can limit its applicability. Indeed, intermediate rewards play a critical role in learning, for example from intrinsic motivation to provide intermediate feedback even in particularly challenging sparse reward tasks. Inspired by this, we propose Generative Augmented Flow Networks (GAFlowNets), a novel learning framework to incorporate intermediate rewards into GFlowNets. We specify intermediate rewards by intrinsic motivation to tackle the exploration problem in sparse reward environments. GAFlowNets can leverage edge-based and state-based intrinsic rewards in a joint way to improve exploration. Based on extensive experiments on the GridWorld task, we demonstrate the effectiveness and efficiency of GAFlowNet in terms of convergence, performance, and diversity of solutions. We further show that GAFlowNet is scalable to a more complex and large-scale molecule generation domain, where it achieves consistent and significant performance improvement.

Multi-user delay constrained scheduling is important in many real-world applications including wireless communication, live streaming, and cloud computing. Yet, it poses a critical challenge since the scheduler needs to make real-time decisions to guarantee the delay and resource constraints simultaneously without prior information of system dynamics, which can be time-varying and hard to estimate. Moreover, many practical scenarios suffer from partial observability issues, e.g., due to sensing noise or hidden correlation. To tackle these challenges, we propose a deep reinforcement learning (DRL) algorithm, named Recurrent Softmax Delayed Deep Double Deterministic Policy Gradient ($\mathtt{RSD4}$), which is a data-driven method based on a Partially Observed Markov Decision Process (POMDP) formulation. $\mathtt{RSD4}$ guarantees resource and delay constraints by Lagrangian dual and delay-sensitive queues, respectively. It also efficiently tackles partial observability with a memory mechanism enabled by the recurrent neural network (RNN) and introduces user-level decomposition and node-level merging to ensure scalability. Extensive experiments on simulated/real-world datasets demonstrate that $\mathtt{RSD4}$ is robust to system dynamics and partially observable environments, and achieves superior performances over existing DRL and non-DRL-based methods., Comment: Accepted by MobiHoc 2022

We study reinforcement learning with linear function approximation where the transition probability and reward functions are linear with respect to a feature mapping $\boldsymbol{\phi}(s,a)$. Specifically, we consider the episodic inhomogeneous linear Markov Decision Process (MDP), and propose a novel computation-efficient algorithm, LSVI-UCB$^+$, which achieves an $\widetilde{O}(Hd\sqrt{T})$ regret bound where $H$ is the episode length, $d$ is the feature dimension, and $T$ is the number of steps. LSVI-UCB$^+$ builds on weighted ridge regression and upper confidence value iteration with a Bernstein-type exploration bonus. Our statistical results are obtained with novel analytical tools, including a new Bernstein self-normalized bound with conservatism on elliptical potentials, and refined analysis of the correction term. This is a minimax optimal algorithm for linear MDPs up to logarithmic factors, which closes the $\sqrt{Hd}$ gap between the upper bound of $\widetilde{O}(\sqrt{H^3d^3T})$ in (Jin et al., 2020) and lower bound of $\Omega(Hd\sqrt{T})$ for linear MDPs., Comment: This is an updated version of our ICML camera-ready version, which has a technical error in building the over-optimistic value function. In this version, this error is fixed using the technique of the "rare-switching" value function from (He et al., 2022)

Despite the superior empirical success of deep meta-learning, theoretical understanding of overparameterized meta-learning is still limited. This paper studies the generalization of a widely used meta-learning approach, Model-Agnostic Meta-Learning (MAML), which aims to find a good initialization for fast adaptation to new tasks. Under a mixed linear regression model, we analyze the generalization properties of MAML trained with SGD in the overparameterized regime. We provide both upper and lower bounds for the excess risk of MAML, which captures how SGD dynamics affect these generalization bounds. With such sharp characterizations, we further explore how various learning parameters impact the generalization capability of overparameterized MAML, including explicitly identifying typical data and task distributions that can achieve diminishing generalization error with overparameterization, and characterizing the impact of adaptation learning rate on both excess risk and the early stopping time. Our theoretical findings are further validated by experiments., Comment: 45 pages, 3 figures

In this paper, we study a novel episodic risk-sensitive Reinforcement Learning (RL) problem, named Iterated CVaR RL, which aims to maximize the tail of the reward-to-go at each step, and focuses on tightly controlling the risk of getting into catastrophic situations at each stage. This formulation is applicable to real-world tasks that demand strong risk avoidance throughout the decision process, such as autonomous driving, clinical treatment planning and robotics. We investigate two performance metrics under Iterated CVaR RL, i.e., Regret Minimization and Best Policy Identification. For both metrics, we design efficient algorithms ICVaR-RM and ICVaR-BPI, respectively, and provide nearly matching upper and lower bounds with respect to the number of episodes $K$. We also investigate an interesting limiting case of Iterated CVaR RL, called Worst Path RL, where the objective becomes to maximize the minimum possible cumulative reward. For Worst Path RL, we propose an efficient algorithm with constant upper and lower bounds. Finally, our techniques for bounding the change of CVaR due to the value function shift and decomposing the regret via a distorted visitation distribution are novel, and can find applications in other risk-sensitive RL problems.

Training deep reinforcement learning (DRL) models usually requires high computation costs. Therefore, compressing DRL models possesses immense potential for training acceleration and model deployment. However, existing methods that generate small models mainly adopt the knowledge distillation-based approach by iteratively training a dense network. As a result, the training process still demands massive computing resources. Indeed, sparse training from scratch in DRL has not been well explored and is particularly challenging due to non-stationarity in bootstrap training. In this work, we propose a novel sparse DRL training framework, "the Rigged Reinforcement Learning Lottery" (RLx2), which builds upon gradient-based topology evolution and is capable of training a sparse DRL model based entirely on a sparse network. Specifically, RLx2 introduces a novel multi-step TD target mechanism with a dynamic-capacity replay buffer to achieve robust value learning and efficient topology exploration in sparse models. It also reaches state-of-the-art sparse training performance in several tasks, showing 7.5\times-20\times model compression with less than 3% performance degradation and up to 20\times and 50\times FLOPs reduction for training and inference, respectively., Comment: ICLR 2023 spotlight

Topology impacts important network performance metrics, including link utilization, throughput and latency, and is of central importance to network operators. However, due to the combinatorial nature of network topology, it is extremely difficult to obtain an optimal solution, especially since topology planning in networks also often comes with management-specific constraints. As a result, local optimization with hand-tuned heuristic methods from human experts are often adopted in practice. Yet, heuristic methods cannot cover the global topology design space while taking into account constraints, and cannot guarantee to find good solutions. In this paper, we propose a novel deep reinforcement learning (DRL) algorithm, called Advantage Actor Critic-Graph Searching (A2C-GS), for network topology optimization. A2C-GS consists of three novel components, including a verifier to validate the correctness of a generated network topology, a graph neural network (GNN) to efficiently approximate topology rating, and a DRL actor layer to conduct a topology search. A2C-GS can efficiently search over large topology space and output topology with satisfying performance. We conduct a case study based on a real network scenario, and our experimental results demonstrate the superior performance of A2C-GS in terms of both efficiency and performance.

Despite the remarkable success of deep multi-modal learning in practice, it has not been well-explained in theory. Recently, it has been observed that the best uni-modal network outperforms the jointly trained multi-modal network, which is counter-intuitive since multiple signals generally bring more information. This work provides a theoretical explanation for the emergence of such performance gap in neural networks for the prevalent joint training framework. Based on a simplified data distribution that captures the realistic property of multi-modal data, we prove that for the multi-modal late-fusion network with (smoothed) ReLU activation trained jointly by gradient descent, different modalities will compete with each other. The encoder networks will learn only a subset of modalities. We refer to this phenomenon as modality competition. The losing modalities, which fail to be discovered, are the origins where the sub-optimality of joint training comes from. Experimentally, we illustrate that modality competition matches the intrinsic behavior of late-fusion joint training., Comment: 41 pages, 2 figures

In this paper, we generalize the concept of heavy-tailed multi-armed bandits to adversarial environments, and develop robust best-of-both-worlds algorithms for heavy-tailed multi-armed bandits (MAB), where losses have $\alpha$-th ($1<\alpha\le 2$) moments bounded by $\sigma^\alpha$, while the variances may not exist. Specifically, we design an algorithm \texttt{HTINF}, when the heavy-tail parameters $\alpha$ and $\sigma$ are known to the agent, \texttt{HTINF} simultaneously achieves the optimal regret for both stochastic and adversarial environments, without knowing the actual environment type a-priori. When $\alpha,\sigma$ are unknown, \texttt{HTINF} achieves a $\log T$-style instance-dependent regret in stochastic cases and $o(T)$ no-regret guarantee in adversarial cases. We further develop an algorithm \texttt{AdaTINF}, achieving $\mathcal O(\sigma K^{1-\nicefrac 1\alpha}T^{\nicefrac{1}{\alpha}})$ minimax optimal regret even in adversarial settings, without prior knowledge on $\alpha$ and $\sigma$. This result matches the known regret lower-bound (Bubeck et al., 2013), which assumed a stochastic environment and $\alpha$ and $\sigma$ are both known. To our knowledge, the proposed \texttt{HTINF} algorithm is the first to enjoy a best-of-both-worlds regret guarantee, and \texttt{AdaTINF} is the first algorithm that can adapt to both $\alpha$ and $\sigma$ to achieve optimal gap-indepedent regret bound in classical heavy-tailed stochastic MAB setting and our novel adversarial formulation.

Conservatism has led to significant progress in offline reinforcement learning (RL) where an agent learns from pre-collected datasets. However, as many real-world scenarios involve interaction among multiple agents, it is important to resolve offline RL in the multi-agent setting. Given the recent success of transferring online RL algorithms to the multi-agent setting, one may expect that offline RL algorithms will also transfer to multi-agent settings directly. Surprisingly, we empirically observe that conservative offline RL algorithms do not work well in the multi-agent setting -- the performance degrades significantly with an increasing number of agents. Towards mitigating the degradation, we identify a key issue that non-concavity of the value function makes the policy gradient improvements prone to local optima. Multiple agents exacerbate the problem severely, since the suboptimal policy by any agent can lead to uncoordinated global failure. Following this intuition, we propose a simple yet effective method, Offline Multi-Agent RL with Actor Rectification (OMAR), which combines the first-order policy gradients and zeroth-order optimization methods to better optimize the conservative value functions over the actor parameters. Despite the simplicity, OMAR achieves state-of-the-art results in a variety of multi-agent control tasks.

In this paper, we develop a novel virtual-queue-based online algorithm for online convex optimization (OCO) problems with long-term and time-varying constraints and conduct a performance analysis with respect to the dynamic regret and constraint violations. We design a new update rule of dual variables and a new way of incorporating time-varying constraint functions into the dual variables. To the best of our knowledge, our algorithm is the first parameter-free algorithm to simultaneously achieve sublinear dynamic regret and constraint violations. Our proposed algorithm also outperforms the state-of-the-art results in many aspects, e.g., our algorithm does not require the Slater condition. Meanwhile, for a group of practical and widely-studied constrained OCO problems in which the variation of consecutive constraints is smooth enough across time, our algorithm achieves $O(1)$ constraint violations. Furthermore, we extend our algorithm and analysis to the case when the time horizon $T$ is unknown. Finally, numerical experiments are conducted to validate the theoretical guarantees of our algorithm, and some applications of our proposed framework will be outlined., Comment: 31 pages, it has been accepted at Performance 2021

In this paper, we formulate a Collaborative Pure Exploration in Kernel Bandit problem (CoPE-KB), which provides a novel model for multi-agent multi-task decision making under limited communication and general reward functions, and is applicable to many online learning tasks, e.g., recommendation systems and network scheduling. We consider two settings of CoPE-KB, i.e., Fixed-Confidence (FC) and Fixed-Budget (FB), and design two optimal algorithms CoopKernelFC (for FC) and CoopKernelFB (for FB). Our algorithms are equipped with innovative and efficient kernelized estimators to simultaneously achieve computation and communication efficiency. Matching upper and lower bounds under both the statistical and communication metrics are established to demonstrate the optimality of our algorithms. The theoretical bounds successfully quantify the influences of task similarities on learning acceleration and only depend on the effective dimension of the kernelized feature space. Our analytical techniques, including data dimension decomposition, linear structured instance transformation and (communication) round-speedup induction, are novel and applicable to other bandit problems. Empirical evaluations are provided to validate our theoretical results and demonstrate the performance superiority of our algorithms.

We consider the Scale-Free Adversarial Multi-Armed Bandit (MAB) problem with unrestricted feedback delays. In contrast to the standard assumption that all losses are $[0,1]$-bounded, in our setting, losses can fall in a general bounded interval $[-L, L]$, unknown to the agent beforehand. Furthermore, the feedback of each arm pull can experience arbitrary delays. We propose a novel approach named Scale-Free Delayed INF (SFD-INF) for this novel setting, which combines a recent "convex combination trick" together with a novel doubling and skipping technique. We then present two instances of SFD-INF, each with carefully designed delay-adapted learning scales. The first one SFD-TINF uses $\frac 12$-Tsallis entropy regularizer and can achieve $\widetilde{\mathcal O}(\sqrt{K(D+T)}L)$ regret when the losses are non-negative, where $K$ is the number of actions, $T$ is the number of steps, and $D$ is the total feedback delay. This bound nearly matches the $\Omega((\sqrt{KT}+\sqrt{D\log K})L)$ lower-bound when regarding $K$ as a constant independent of $T$. The second one, SFD-LBINF, works for general scale-free losses and achieves a small-loss style adaptive regret bound $\widetilde{\mathcal O}(\sqrt{K\mathbb{E}[\tilde{\mathfrak L}_T^2]}+\sqrt{KDL})$, which falls to the $\widetilde{\mathcal O}(\sqrt{K(D+T)}L)$ regret in the worst case and is thus more general than SFD-TINF despite a more complicated analysis and several extra logarithmic dependencies. Moreover, both instances also outperform the existing algorithms for non-delayed (i.e., $D=0$) scale-free adversarial MAB problems, which can be of independent interest., Comment: Preliminary work, merged to arXiv:2301.10500

We propose Banker-OMD, a novel framework generalizing the classical Online Mirror Descent (OMD) technique in online learning algorithm design. Banker-OMD allows algorithms to robustly handle delayed feedback, and offers a general methodology for achieving $\tilde{O}(\sqrt{T} + \sqrt{D})$-style regret bounds in various delayed-feedback online learning tasks, where $T$ is the time horizon length and $D$ is the total feedback delay. We demonstrate the power of Banker-OMD with applications to three important bandit scenarios with delayed feedback, including delayed adversarial Multi-armed bandits (MAB), delayed adversarial linear bandits, and a novel delayed best-of-both-worlds MAB setting. Banker-OMD achieves nearly-optimal performance in all the three settings. In particular, it leads to the first delayed adversarial linear bandit algorithm achieving $\tilde{O}(\text{poly}(n)(\sqrt{T} + \sqrt{D}))$ regret., Comment: Preliminary work, merged to arXiv:2301.10500

The world provides us with data of multiple modalities. Intuitively, models fusing data from different modalities outperform their uni-modal counterparts, since more information is aggregated. Recently, joining the success of deep learning, there is an influential line of work on deep multi-modal learning, which has remarkable empirical results on various applications. However, theoretical justifications in this field are notably lacking. Can multi-modal learning provably perform better than uni-modal? In this paper, we answer this question under a most popular multi-modal fusion framework, which firstly encodes features from different modalities into a common latent space and seamlessly maps the latent representations into the task space. We prove that learning with multiple modalities achieves a smaller population risk than only using its subset of modalities. The main intuition is that the former has a more accurate estimate of the latent space representation. To the best of our knowledge, this is the first theoretical treatment to capture important qualitative phenomena observed in real multi-modal applications from the generalization perspective. Combining with experiment results, we show that multi-modal learning does possess an appealing formal guarantee., Comment: Accepted to NeurIPS 2021

Federated Learning (FL) coordinates with numerous heterogeneous devices to collaboratively train a shared model while preserving user privacy. Despite its multiple advantages, FL faces new challenges. One challenge arises when devices drop out of the training process beyond the control of the central server. In this case, the convergence of popular FL algorithms such as FedAvg is severely influenced by the straggling devices. To tackle this challenge, we study federated learning algorithms under arbitrary device unavailability and propose an algorithm named Memory-augmented Impatient Federated Averaging (MIFA). Our algorithm efficiently avoids excessive latency induced by inactive devices, and corrects the gradient bias using the memorized latest updates from the devices. We prove that MIFA achieves minimax optimal convergence rates on non-i.i.d. data for both strongly convex and non-convex smooth functions. We also provide an explicit characterization of the improvement over baseline algorithms through a case study, and validate the results by numerical experiments on real-world datasets.

We consider the best-of-both-worlds problem for learning an episodic Markov Decision Process through $T$ episodes, with the goal of achieving $\widetilde{\mathcal{O}}(\sqrt{T})$ regret when the losses are adversarial and simultaneously $\mathcal{O}(\text{polylog}(T))$ regret when the losses are (almost) stochastic. Recent work by [Jin and Luo, 2020] achieves this goal when the fixed transition is known, and leaves the case of unknown transition as a major open question. In this work, we resolve this open problem by using the same Follow-the-Regularized-Leader ($\text{FTRL}$) framework together with a set of new techniques. Specifically, we first propose a loss-shifting trick in the $\text{FTRL}$ analysis, which greatly simplifies the approach of [Jin and Luo, 2020] and already improves their results for the known transition case. Then, we extend this idea to the unknown transition case and develop a novel analysis which upper bounds the transition estimation error by (a fraction of) the regret itself in the stochastic setting, a key property to ensure $\mathcal{O}(\text{polylog}(T))$ regret., Comment: The camera-ready version for NeurIPS 2021

Tackling overestimation in $Q$-learning is an important problem that has been extensively studied in single-agent reinforcement learning, but has received comparatively little attention in the multi-agent setting. In this work, we empirically demonstrate that QMIX, a popular $Q$-learning algorithm for cooperative multi-agent reinforcement learning (MARL), suffers from a more severe overestimation in practice than previously acknowledged, and is not mitigated by existing approaches. We rectify this with a novel regularization-based update scheme that penalizes large joint action-values that deviate from a baseline and demonstrate its effectiveness in stabilizing learning. Furthermore, we propose to employ a softmax operator, which we efficiently approximate in a novel way in the multi-agent setting, to further reduce the potential overestimation bias. Our approach, Regularized Softmax (RES) Deep Multi-Agent $Q$-Learning, is general and can be applied to any $Q$-learning based MARL algorithm. We demonstrate that, when applied to QMIX, RES avoids severe overestimation and significantly improves performance, yielding state-of-the-art results in a variety of cooperative multi-agent tasks, including the challenging StarCraft II micromanagement benchmarks.

Existing risk-aware multi-armed bandit models typically focus on risk measures of individual options such as variance. As a result, they cannot be directly applied to important real-world online decision making problems with correlated options. In this paper, we propose a novel Continuous Mean-Covariance Bandit (CMCB) model to explicitly take into account option correlation. Specifically, in CMCB, there is a learner who sequentially chooses weight vectors on given options and observes random feedback according to the decisions. The agent's objective is to achieve the best trade-off between reward and risk, measured with option covariance. To capture different reward observation scenarios in practice, we consider three feedback settings, i.e., full-information, semi-bandit and full-bandit feedback. We propose novel algorithms with optimal regrets (within logarithmic factors), and provide matching lower bounds to validate their optimalities. The experimental results also demonstrate the superiority of our algorithms. To the best of our knowledge, this is the first work that considers option correlation in risk-aware bandits and explicitly quantifies how arbitrary covariance structures impact the learning performance. The novel analytical techniques we developed for exploiting the estimated covariance to build concentration and bounding the risk of selected actions based on sampling strategy properties can likely find applications in other bandit analysis and be of independent interests.

In this paper, we study a family of conservative bandit problems (CBPs) with sample-path reward constraints, i.e., the learner's reward performance must be at least as well as a given baseline at any time. We propose a One-Size-Fits-All solution to CBPs and present its applications to three encompassed problems, i.e. conservative multi-armed bandits (CMAB), conservative linear bandits (CLB) and conservative contextual combinatorial bandits (CCCB). Different from previous works which consider high probability constraints on the expected reward, we focus on a sample-path constraint on the actually received reward, and achieve better theoretical guarantees ($T$-independent additive regrets instead of $T$-dependent) and empirical performance. Furthermore, we extend the results and consider a novel conservative mean-variance bandit problem (MV-CBP), which measures the learning performance with both the expected reward and variability. For this extended problem, we provide a novel algorithm with $O(1/T)$ normalized additive regrets ($T$-independent in the cumulative form) and validate this result through empirical evaluation.

We study the multi-armed bandit (MAB) problem with composite and anonymous feedback. In this model, the reward of pulling an arm spreads over a period of time (we call this period as reward interval) and the player receives partial rewards of the action, convoluted with rewards from pulling other arms, successively. Existing results on this model require prior knowledge about the reward interval size as an input to their algorithms. In this paper, we propose adaptive algorithms for both the stochastic and the adversarial cases, without requiring any prior information about the reward interval. For the stochastic case, we prove that our algorithm guarantees a regret that matches the lower bounds (in order). For the adversarial case, we propose the first algorithm to jointly handle non-oblivious adversary and unknown reward interval size. We also conduct simulations based on real-world dataset. The results show that our algorithms outperform existing benchmarks.

We study the online restless bandit problem, where the state of each arm evolves according to a Markov chain, and the reward of pulling an arm depends on both the pulled arm and the current state of the corresponding Markov chain. In this paper, we propose Restless-UCB, a learning policy that follows the explore-then-commit framework. In Restless-UCB, we present a novel method to construct offline instances, which only requires $O(N)$ time-complexity ($N$ is the number of arms) and is exponentially better than the complexity of existing learning policy. We also prove that Restless-UCB achieves a regret upper bound of $\tilde{O}((N+M^3)T^{2\over 3})$, where $M$ is the Markov chain state space size and $T$ is the time horizon. Compared to existing algorithms, our result eliminates the exponential factor (in $M,N$) in the regret upper bound, due to a novel exploitation of the sparsity in transitions in general restless bandit problems. As a result, our analysis technique can also be adopted to tighten the regret bounds of existing algorithms. Finally, we conduct experiments based on real-world dataset, to compare the Restless-UCB policy with state-of-the-art benchmarks. Our results show that Restless-UCB outperforms existing algorithms in regret, and significantly reduces the running time.