论文检索

Deep Learning · Large Language Models

Xing Li, Zeyu Xing, Yiming Li, Linping Qu, Huiling Zhen, Yiwu Yao, Wulong Liu, Sinno Jialin Pan, Mingxuan Yuan

KV cache quantization can improve Large Language Models (LLMs) inference throughput and latency in long contexts and large batch-size scenarios while preserving LLMs effectiveness.However, current methods have three unsolved issues: overlooking layer-wise sensitivity to KV cache quantization, high overhead of online fine-grained decision-making, and low flexibility to different LLMs and constraints. Therefore, we theoretically analyze the inherent correlation of layer-wise transformer attention patterns to KV cache quantization errors and study why key cache is generally more important than value cache for quantization error reduction. We further propose a simple yet effective framework KVTuner to adaptively search for the optimal hardware-friendly layer-wise KV quantization precision pairs for coarse-grained KV cache with multi-objective optimization and directly utilize the offline searched configurations during online inference.To reduce the computational cost of offline calibration, we utilize the intra-layer KV precision pair pruning and inter-layer clustering to reduce the search space. Experimental results show that we can achieve nearly lossless 3.25-bit mixed precision KV cache quantization for LLMs like Llama-3.1-8B-Instruct and 4.0-bit for sensitive models like Qwen2.5-7B-Instruct on mathematical reasoning tasks. The maximum inference throughput can be improved by 21.25\% compared with KIVI-KV8 quantization over various context lengths. Our code and searched configurations are available at https://github.com/cmd2001/KVTuner.

Miscellaneous Aspects of Machine Learning · Online Learning, Active Learning and Bandits

Giuseppe Serra, Florian Buettner

Given the ability to model more realistic and dynamic problems, Federated Continual Learning (FCL) has been increasingly investigated recently. A well-known problem encountered in this setting is the so-called catastrophic forgetting, for which the learning model is inclined to focus on more recent tasks while forgetting the previously learned knowledge. The majority of the current approaches in FCL propose generative-based solutions to solve said problem. However, this setting requires multiple training epochs over the data, implying an offline setting where datasets are stored locally and remain unchanged over time. Furthermore, the proposed solutions are tailored for vision tasks solely. To overcome these limitations, we propose a new approach to deal with different modalities in the online scenario where new data arrive in streams of mini-batches that can only be processed once. To solve catastrophic forgetting, we propose an uncertainty-aware memory-based approach. Specifically, we suggest using an estimator based on the Bregman Information (BI) to compute the model's variance at the sample level. Through measures of predictive uncertainty, we retrieve samples with specific characteristics, and – by retraining the model on such samples – we demonstrate the potential of this approach to reduce the forgetting effect in realistic settings while maintaining data confidentiality and competitive communication efficiency compared to state-of-the-art approaches.

Ioannis Maniadis Metaxas, Georgios Tzimiropoulos, ioannis Patras

Self-supervised learning has recently emerged as the preeminent pretraining paradigm across and between modalities, with remarkable results. In the image domain specifically, group (or cluster) discrimination has been one of the most successful methods. However, such frameworks need to guard against heavily imbalanced cluster assignments to prevent collapse to trivial solutions. Existing works typically solve this by reweighing cluster assignments to promote balance, or with offline operations (e.g. regular re-clustering) that prevent collapse. However, the former typically requires large batch sizes, which leads to increased resource requirements, and the latter introduces scalability issues with regard to large datasets. In this work, we propose ExCB, a framework that tackles this problem with a novel cluster balancing method. ExCB estimates the relative size of the clusters across batches and balances them by adjusting cluster assignments, proportionately to their relative size and in an online manner. Thereby, it overcomes previous methods' dependence on large batch sizes and is fully online, and therefore scalable to any dataset. We conduct extensive experiments to evaluate our approach and demonstrate that ExCB: a) achieves state of the art results with significantly reduced resource requirements compared to previous works, b) is fully online, and therefore scalable to large datasets, and c) is stable and effective even with very small batch sizes.

Vincent Liu, James R. Wright, Mrtha White

Offline reinforcement learning—learning a policy from a batch of data—is known to be hard for general MDPs. These results motivate the need to look at specific classes of MDPs where offline reinforcement learning might be feasible. In this work, we explore a restricted class of MDPs to obtain guarantees for offline reinforcement learning. The key property, which we call Action Impact Regularity (AIR), is that actions primarily impact a part of the state (an endogenous component) and have limited impact on the remaining part of the state (an exogenous component). AIR is a strong assumption, but it nonetheless holds in a number of real-world domains including financial markets. We discuss algorithms that exploit the AIR property, and provide a theoretical analysis for an algorithm based on Fitted-Q Iteration. Finally, we demonstrate that the algorithm outperforms existing offline reinforcement learning algorithms across different data collection policies in simulated and real world environments where the regularity holds.

Louis Mahon, Thomas Lukasiewicz

Online deep clustering refers to the joint use of a feature extraction network and a clustering model to assign cluster labels to each new data point or batch as it is processed. While faster and more versatile than offline methods, online clustering can easily reach the collapsed solution where the encoder maps all inputs to the same point and all are put into a single cluster. Successful existing models have employed various techniques to avoid this problem, most of which require data augmentation or which aim to make the average soft assignment across the dataset the same for each cluster. We propose a method that does not require data augmentation, and that, differently from existing methods, regularizes the hard assignments. Using a Bayesian framework, we derive an intuitive optimization objective that can be straightforwardly included in the training of the encoder network. Tested on four image datasets, it consistently avoids collapse more robustly than other methods and leads to more accurate clustering. We also conduct further experiments and analyses justifying our choice to regularize the hard cluster assignments. Code is available at https://github.com/Lou1sM/online_hard_clustering.

Hongbo Zhang, Guang Wang, Xu Wang, Zhengyang Zhou, Chen Zhang, Zheng Dong, Yang Wang

One of the most important tasks in ride-hailing is order dispatching, i.e., assigning unserved orders to available drivers. Recent order dispatching has achieved a significant improvement due to the advance of reinforcement learning, which has been approved to be able to effectively address sequential decision-making problems like order dispatching. However, most existing reinforcement learning methods require agents to learn the optimal policy by interacting with environments online, which is challenging or impractical for real-world deployment due to high costs or safety concerns. For example, due to the spatiotemporally unbalanced supply and demand, online reinforcement learning-based order dispatching may significantly impact the revenue of the ride-hailing platform and passenger experience during the policy learning period. Hence, in this work, we develop an offline deep reinforcement learning framework called NondBREM for large-scale order dispatching, which learns policy from only the accumulated logged data to avoid costly and unsafe interactions with the environment. In NondBREM, a Nondeterministic Batch-Constrained Q-learning (NondBCQ) module is developed to reduce the algorithm extrapolation error and a Random Ensemble Mixture (REM) module that integrates multiple value networks with multi-head networks is utilized to improve the model generalization and robustness. Extensive experiments on large-scale real-world ride-hailing datasets show the superiority of our design.

Han Guo, William Brandon, Radostin Cholakov, Jonathan Ragan-Kelley, Eric P. Xing, Yoon Kim

The deployment of large language models (LLMs) is often constrained by memory bandwidth, where the primary bottleneck is the cost of transferring model parameters from the GPU’s global memory to its registers. When coupled with custom kernels that fuse the dequantization and matmul operations, weight-only quantization can thus enable faster inference by reducing the amount of memory movement. However, developing high-performance kernels for weight-quantized LLMs presents substantial challenges, especially when the weights are compressed to non-evenly-divisible bit widths (e.g., 3 bits) with non-uniform, lookup table (LUT) quantization. This paper describes FLUTE, a flexible lookup table engine for LUT-quantized LLMs, which uses offline restructuring of the quantized weight matrix to minimize bit manipulations associated with unpacking, and vectorization and duplication of the lookup table to mitigate shared memory bandwidth constraints. At batch sizes < 32 and quantization group size of 128 (typical in LLM inference), the FLUTE kernel can be 2-4x faster than existing GEMM kernels. As an application of FLUTE, we explore a simple extension to lookup table-based NormalFloat quantization and apply it to quantize LLaMA3 to various configurations, obtaining competitive quantization performance against strong baselines while obtaining an end-to-end throughput increase of 1.5 to 2 times.

Reinforcement Learning · Batch Offline

Junghyuk Yeom, Yonghyeon Jo, Jeongmo Kim, Sanghyeon Lee, Seungyul Han

Constraint-based offline reinforcement learning (RL) involves policy constraints or imposing penalties on the value function to mitigate overestimation errors caused by distributional shift. This paper focuses on a limitation in existing offline RL methods with penalized value function, indicating the potential for underestimation bias due to unnecessary bias introduced in the value function. To address this concern, we propose Exclusively Penalized Q-learning (EPQ), which reduces estimation bias in the value function by selectively penalizing states that are prone to inducing estimation errors. Numerical results show that our method significantly reduces underestimation bias and improves performance in various offline control tasks compared to other offline RL methods.

Reinforcement Learning · Batch Offline

Abdullah Akgül, Manuel Haussmann, Melih Kandemir

Current approaches to model-based offline reinforcement learning often incorporate uncertainty-based reward penalization to address the distributional shift problem. These approaches, commonly known as pessimistic value iteration, use Monte Carlo sampling to estimate the Bellman target to perform temporal difference-based policy evaluation. We find out that the randomness caused by this sampling step significantly delays convergence. We present a theoretical result demonstrating the strong dependency of suboptimality on the number of Monte Carlo samples taken per Bellman target calculation. Our main contribution is a deterministic approximation to the Bellman target that uses progressive moment matching, a method developed originally for deterministic variational inference. The resulting algorithm, which we call Moment Matching Offline Model-Based Policy Optimization (MOMBO), propagates the uncertainty of the next state through a nonlinear Q-network in a deterministic fashion by approximating the distributions of hidden layer activations by a normal distribution. We show that it is possible to provide tighter guarantees for the suboptimality of MOMBO than the existing Monte Carlo sampling approaches. We also observe MOMBO to converge faster than these approaches in a large set of benchmark tasks.

Reinforcement Learning · Batch Offline

Lanqing Li, Hai Zhang, Xinyu Zhang, Shatong Zhu, Yang YU, Junqiao Zhao, Pheng-Ann Heng

As a marriage between offline RL and meta-RL, the advent of offline meta-reinforcement learning (OMRL) has shown great promise in enabling RL agents to multi-task and quickly adapt while acquiring knowledge safely. Among which, context-based OMRL (COMRL) as a popular paradigm, aims to learn a universal policy conditioned on effective task representations. In this work, by examining several key milestones in the field of COMRL, we propose to integrate these seemingly independent methodologies into a unified framework. Most importantly, we show that the pre-existing COMRL algorithms are essentially optimizing the same mutual information objective between the task variable $M$ and its latent representation $Z$ by implementing various approximate bounds. Such theoretical insight offers ample design freedom for novel algorithms. As demonstrations, we propose a supervised and a self-supervised implementation of $I(Z; M)$, and empirically show that the corresponding optimization algorithms exhibit remarkable generalization across a broad spectrum of RL benchmarks, context shift scenarios, data qualities and deep learning architectures. This work lays the information theoretic foundation for COMRL methods, leading to a better understanding of task representation learning in the context of reinforcement learning. Given itsgenerality, we envision our framework as a promising offline pre-training paradigm of foundation models for decision making.

Reinforcement Learning · Batch Offline

Yihang Yao, Zhepeng Cen, Wenhao Ding, Haohong Lin, Shiqi Liu, Tingnan Zhang, Wenhao Yu, DING ZHAO

Offline safe reinforcement learning (RL) aims to train a policy that satisfies con- straints using a pre-collected dataset. Most current methods struggle with the mismatch between imperfect demonstrations and the desired safe and rewarding performance. In this paper, we mitigate this issue from a data-centric perspective and introduce OASIS (cOnditionAl diStributIon Shaping), a new paradigm in offline safe RL designed to overcome these critical limitations. OASIS utilizes a conditional diffusion model to synthesize offline datasets, thus shaping the data dis- tribution toward a beneficial target domain. Our approach makes compliance with safety constraints through effective data utilization and regularization techniques to benefit offline safe RL training. Comprehensive evaluations on public benchmarks and varying datasets showcase OASIS’s superiority in benefiting offline safe RL agents to achieve high-reward behavior while satisfying the safety constraints, out- performing established baselines. Furthermore, OASIS exhibits high data efficiency and robustness, making it suitable for real-world applications, particularly in tasks where safety is imperative and high-quality demonstrations are scarce. More details are available at the website https://sites.google.com/view/saferl-oasis/home.

Reinforcement Learning · Batch Offline

Jaewoo Lee, Sujin Yun, Taeyoung Yun, Jinkyoo Park

Offline Reinforcement Learning (Offline RL) presents challenges of learning effective decision-making policies from static datasets without any online interactions. Data augmentation techniques, such as noise injection and data synthesizing, aim to improve Q-function approximation by smoothing the learned state-action region. However, these methods often fall short of directly improving the quality of offline datasets, leading to suboptimal results. In response, we introduce GTA, Generative Trajectory Augmentation, a novel generative data augmentation approach designed to enrich offline data by augmenting trajectories to be both high-rewarding and dynamically plausible. GTA applies a diffusion model within the data augmentation framework. GTA partially noises original trajectories and then denoises them with classifier-free guidance via conditioning on amplified return value. Our results show that GTA, as a general data augmentation strategy, enhances the performance of widely used offline RL algorithms across various tasks with unique challenges. Furthermore, we conduct a quality analysis of data augmented by GTA and demonstrate that GTA improves the quality of the data. Our code is available at https://github.com/Jaewoopudding/GTA

Reinforcement Learning · Batch Offline

Seohong Park, Kevin Frans, Sergey Levine, Aviral Kumar

While imitation learning requires access to high-quality data, offline reinforcement learning (RL) should, in principle, perform similarly or better with substantially lower data quality by using a value function. However, current results indicate that offline RL often performs worse than imitation learning, and it is often unclear what holds back the performance of offline RL. Motivated by this observation, we aim to understand the bottlenecks in current offline RL algorithms. While poor performance of offline RL is typically attributed to an imperfect value function, we ask: *is the main bottleneck of offline RL indeed in learning the value function, or something else?* To answer this question, we perform a systematic empirical study of (1) value learning, (2) policy extraction, and (3) policy generalization in offline RL problems, analyzing how these components affect performance. We make two surprising observations. First, we find that the choice of a policy extraction algorithm significantly affects the performance and scalability of offline RL, often more so than the value learning objective. For instance, we show that common value-weighted behavioral cloning objectives (e.g., AWR) do not fully leverage the learned value function, and switching to behavior-constrained policy gradient objectives (e.g., DDPG+BC) often leads to substantial improvements in performance and scalability. Second, we find that a big barrier to improving offline RL performance is often imperfect policy generalization on test-time states out of the support of the training data, rather than policy learning on in-distribution states. We then show that the use of suboptimal but high-coverage data or test-time policy training techniques can address this generalization issue in practice. Specifically, we propose two simple test-time policy improvement methods and show that these methods lead to better performance.

Reinforcement Learning · Batch Offline

Seokin Seo, Byung-Jun Lee, Jongmin Lee, HyeongJoo Hwang, Hongseok Yang, Kee-Eung Kim

We consider offline imitation learning (IL), which aims to train an agent to imitate from the dataset of expert demonstrations without online interaction with the environment. Behavioral Cloning (BC) has been a simple yet effective approach to offline IL, but it is also well-known to be vulnerable to the covariate shift resulting from the mismatch between the state distributions induced by the learned policy and the expert policy. Moreover, as often occurs in practice, when expert datasets are collected from an arbitrary state distribution instead of a stationary one, these shifts become more pronounced, potentially leading to substantial failures in existing IL methods. Specifically, we focus on covariate shift resulting from arbitrary state data distributions, such as biased data collection or incomplete trajectories, rather than shifts induced by changes in dynamics or noisy expert actions. In this paper, to mitigate the effect of the covariate shifts in BC, we propose DrilDICE, which utilizes a distributionally robust BC objective by employing a stationary distribution correction ratio estimation (DICE) to derive a feasible solution. We evaluate the effectiveness of our method through an extensive set of experiments covering diverse covariate shift scenarios. The results demonstrate the efficacy of the proposed approach in improving the robustness against the shifts, outperforming existing offline IL methods in such scenarios.

Reinforcement Learning · Batch Offline

Sili Huang, Jifeng Hu, Zhejian Yang, Liwei Yang, Tao Luo, Hechang Chen, Lichao Sun, Bo Yang

Recent works have shown the remarkable superiority of transformer models in reinforcement learning (RL), where the decision-making problem is formulated as sequential generation. Transformer-based agents could emerge with self-improvement in online environments by providing task contexts, such as multiple trajectories, called in-context RL. However, due to the quadratic computation complexity of attention in transformers, current in-context RL methods suffer from huge computational costs as the task horizon increases. In contrast, the Mamba model is renowned for its efficient ability to process long-term dependencies, which provides an opportunity for in-context RL to solve tasks that require long-term memory. To this end, we first implement Decision Mamba (DM) by replacing the backbone of Decision Transformer (DT). Then, we propose a Decision Mamba-Hybrid (DM-H) with the merits of transformers and Mamba in high-quality prediction and long-term memory. Specifically, DM-H first generates high-value sub-goals from long-term memory through the Mamba model. Then, we use sub-goals to prompt the transformer, establishing high-quality predictions. Experimental results demonstrate that DM-H achieves state-of-the-art in long and short-term tasks, such as D4RL, Grid World, and Tmaze benchmarks. Regarding efficiency, the online testing of DM-H in the long-term task is 28$\times$ times faster than the transformer-based baselines.

Reinforcement Learning · Batch Offline

Shreyas Chaudhari, Ameet Deshpande, Bruno C. da Silva, Philip Thomas

Evaluating policies using off-policy data is crucial for applying reinforcement learning to real-world problems such as healthcare and autonomous driving. Previous methods for *off-policy evaluation* (OPE) generally suffer from high variance or irreducible bias, leading to unacceptably high prediction errors. In this work, we introduce STAR, a framework for OPE that encompasses a broad range of estimators -- which include existing OPE methods as special cases -- that achieve lower mean squared prediction errors. STAR leverages state abstraction to distill complex, potentially continuous problems into compact, discrete models which we call *abstract reward processes* (ARPs). Predictions from ARPs estimated from off-policy data are provably consistent (asymptotically correct). Rather than proposing a specific estimator, we present a new framework for OPE and empirically demonstrate that estimators within STAR outperform existing methods. The best STAR estimator outperforms baselines in all twelve cases studied, and even the median STAR estimator surpasses the baselines in seven out of the twelve cases.

Reinforcement Learning · Batch Offline

Yixiu Mao, Qi Wang, Chen Chen, Yun Qu, Xiangyang Ji

In offline reinforcement learning (RL), addressing the out-of-distribution (OOD) action issue has been a focus, but we argue that there exists an OOD state issue that also impairs performance yet has been underexplored. Such an issue describes the scenario when the agent encounters states out of the offline dataset during the test phase, leading to uncontrolled behavior and performance degradation. To this end, we propose SCAS, a simple yet effective approach that unifies OOD state correction and OOD action suppression in offline RL. Technically, SCAS achieves value-aware OOD state correction, capable of correcting the agent from OOD states to high-value in-distribution states. Theoretical and empirical results show that SCAS also exhibits the effect of suppressing OOD actions. On standard offline RL benchmarks, SCAS achieves excellent performance without additional hyperparameter tuning. Moreover, benefiting from its OOD state correction feature, SCAS demonstrates enhanced robustness against environmental perturbations.

Reinforcement Learning · Batch Offline

Qin-Wen Luo, Ming-Kun Xie, Yewen Wang, Sheng-Jun Huang

Offline-to-online (O2O) reinforcement learning (RL) provides an effective means of leveraging an offline pre-trained policy as initialization to improve performance rapidly with limited online interactions. Recent studies often design fine-tuning strategies for a specific offline RL method and cannot perform general O2O learning from any offline method. To deal with this problem, we disclose that there are evaluation and improvement mismatches between the offline dataset and the online environment, which hinders the direct application of pre-trained policies to online fine-tuning. In this paper, we propose to handle these two mismatches simultaneously, which aims to achieve general O2O learning from any offline method to any online method. Before online fine-tuning, we re-evaluate the pessimistic critic trained on the offline dataset in an optimistic way and then calibrate the misaligned critic with the reliable offline actor to avoid erroneous update. After obtaining an optimistic and and aligned critic, we perform constrained fine-tuning to combat distribution shift during online learning. We show empirically that the proposed method can achieve stable and efficient performance improvement on multiple simulated tasks when compared to the state-of-the-art methods.

Reinforcement Learning · Batch Offline

Allen Nie, Yash Chandak, Christina Yuan, Anirudhan Badrinath, Yannis Flet-Berliac, Emma Brunskill

Offline policy evaluation (OPE) allows us to evaluate and estimate a new sequential decision-making policy's performance by leveraging historical interaction data collected from other policies. Evaluating a new policy online without a confident estimate of its performance can lead to costly, unsafe, or hazardous outcomes, especially in education and healthcare. Several OPE estimators have been proposed in the last decade, many of which have hyperparameters and require training. Unfortunately, choosing the best OPE algorithm for each task and domain is still unclear. In this paper, we propose a new algorithm that adaptively blends a set of OPE estimators given a dataset without relying on an explicit selection using a statistical procedure. We prove that our estimator is consistent and satisfies several desirable properties for policy evaluation. Additionally, we demonstrate that when compared to alternative approaches, our estimator can be used to select higher-performing policies in healthcare and robotics. Our work contributes to improving ease of use for a general-purpose, estimator-agnostic, off-policy evaluation framework for offline RL.

Reinforcement Learning · Batch Offline

Qian Lin, Zongkai Liu, Danying Mo, Chao Yu

In recent years, significant progress has been made in multi-objective reinforcement learning (RL) research, which aims to balance multiple objectives by incorporating preferences for each objective. In most existing studies, specific preferences must be provided during deployment to indicate the desired policies explicitly. However, designing these preferences depends heavily on human prior knowledge, which is typically obtained through extensive observation of high-performing demonstrations with expected behaviors. In this work, we propose a simple yet effective offline adaptation framework for multi-objective RL problems without assuming handcrafted target preferences, but only given several demonstrations to implicitly indicate the preferences of expected policies. Additionally, we demonstrate that our framework can naturally be extended to meet constraints on safety-critical objectives by utilizing safe demonstrations, even when the safety thresholds are unknown. Empirical results on offline multi-objective and safe tasks demonstrate the capability of our framework to infer policies that align with real preferences while meeting the constraints implied by the provided demonstrations.