Overcoming the Sim-to-Real Gap: Leveraging Simulation to Learn to Explore for Real-World RL

In order to mitigate the sample complexity of real-world reinforcement learning, common practice is to first train a policy in a simulator where samples are cheap, and then deploy this policy in the real world, with the hope that it generalizes effectively. Such \emph{direct sim2real} transfer is not guaranteed to succeed, however, and in cases where it fails, it is unclear how to best utilize the simulator. In this work, we show that in many regimes, while direct sim2real transfer may fail, we can utilize the simulator to learn a set of \emph{exploratory} policies which enable efficient exploration in the real world. In particular, in the setting of low-rank MDPs, we show that coupling these exploratory policies with simple, practical approaches -- least-squares regression oracles and naive randomized exploration -- yields a polynomial sample complexity in the real world, an exponential improvement over direct sim2real transfer, or learning without access to a simulator. To the best of our knowledge, this is the first evidence that simulation transfer yields a provable gain in reinforcement learning in settings where direct sim2real transfer fails. We validate our theoretical results on several realistic robotic simulators and a real-world robotic sim2real task, demonstrating that transferring exploratory policies can yield substantial gains in practice as well.

Cost-Effective Proxy Reward Model Construction with On-Policy and Active Learning

Reinforcement learning with human feedback (RLHF), as a widely adopted approach in current large language model pipelines, is \textit{bottlenecked by the size of human preference data}. While traditional methods rely on offline preference dataset constructions, recent approaches have shifted towards online settings, where a learner uses a small amount of labeled seed data and a large pool of unlabeled prompts to iteratively construct new preference data through self-generated responses and high-quality reward/preference feedback. However, most current online algorithms still focus on preference labeling during policy model updating with given feedback oracles, which incurs significant expert query costs. \textit{We are the first to explore cost-effective proxy reward oracles construction strategies for further labeling preferences or rewards with extremely limited labeled data and expert query budgets}. Our approach introduces two key innovations: (1) on-policy query to avoid OOD and imbalance issues in seed data, and (2) active learning to select the most informative data for preference queries. Using these methods, we train a evaluation model with minimal expert-labeled data, which then effectively labels nine times more preference pairs for further RLHF training. For instance, our model using Direct Preference Optimization (DPO) gains around over 1% average improvement on AlpacaEval2, MMLU-5shot and MMLU-0shot, with only 1.7K query cost. Our methodology is orthogonal to other direct expert query-based strategies and therefore might be integrated with them to further reduce query costs.

Humor in AI: Massive Scale Crowd-Sourced Preferences and Benchmarks for Cartoon Captioning

We present a novel multimodal preference dataset for creative tasks, consisting of over 250 million human ratings on more than 2.2 million captions, collected through crowdsourcing rating data for The New Yorker's weekly cartoon caption contest over the past eight years. This unique dataset supports the development and evaluation of multimodal large language models and preference-based fine-tuning algorithms for humorous caption generation. We propose novel benchmarks for judging the quality of model-generated captions, utilizing both GPT4 and human judgments to establish ranking-based evaluation strategies. Our experimental results highlight the limitations of current fine-tuning methods, such as RLHF and DPO, when applied to creative tasks. Furthermore, we demonstrate that even state-of-the-art models like GPT4 and Claude currently underperform top human contestants in generating humorous captions. As we conclude this extensive data collection effort, we release the entire preference dataset to the research community, fostering further advancements in AI humor generation and evaluation.

Sample Complexity Reduction via Policy Difference Estimation in Tabular Reinforcement Learning

In this paper, we study the non-asymptotic sample complexity for the pure exploration problem in contextual bandits and tabular reinforcement learning (RL): identifying an epsilon-optimal policy from a set of policies with high probability. Existing work in bandits has shown that it is possible to identify the best policy by estimating only the difference between the behaviors of individual policies, which can be substantially cheaper than estimating the behavior of each policy directly. However, the best-known complexities in RL fail to take advantage of this and instead estimate the behavior of each policy directly. Does it suffice to estimate only the differences in the behaviors of policies in RL? We answer this question positively for contextual bandits but in the negative for tabular RL, showing a separation between contextual bandits and RL. However, inspired by this, we show that it almost suffices to estimate only the differences in RL: if we can estimate the behavior of a single reference policy, it suffices to only estimate how any other policy deviates from this reference policy. We develop an algorithm which instantiates this principle and obtains, to the best of our knowledge, the tightest known bound on the sample complexity of tabular RL.

CLIPLoss and Norm-Based Data Selection Methods for Multimodal Contrastive Learning

Data selection has emerged as a core issue for large-scale visual-language model pretaining (e.g., CLIP), particularly with noisy web-curated datasets. Three main data selection approaches are: (1) leveraging external non-CLIP models to aid data selection, (2) training new CLIP-style embedding models that are more effective at selecting high-quality data than the original OpenAI CLIP model, and (3) designing better metrics or strategies universally applicable to any CLIP embedding without requiring specific model properties (e.g., CLIPScore is one popular metric). While the first two approaches have been extensively studied, the third remains under-explored. In this paper, we advance the third approach by proposing two new methods. Firstly, instead of classical CLIP scores that only consider the alignment between two modalities from a single sample, we introduce negCLIPLoss, a CLIP loss-inspired method that adds the alignment between one sample and its contrastive pairs as an extra normalization term for better quality measurement. Secondly, when downstream tasks are known, we propose a new norm-based metric, NormSim, to measure the similarity between pretraining data and target data. We test our methods on the data selection benchmark, DataComp~\cite{gadre2023datacomp}. Compared to the best baseline using only OpenAI's CLIP-L/14, our methods achieve a 5.3\% improvement on ImageNet-1k and a 2.8\% improvement on 38 downstream evaluation tasks. Moreover, both negCLIPLoss and NormSim are compatible with existing techniques. By combining our methods with the current best methods DFN~\cite{fang2023data} and HYPE~\cite{kim2024hype}, we can boost average performance on downstream tasks by 0.9\%, achieving a new state-of-the-art.

Variance Alignment Score: A Simple But Tough-to-Beat Data Selection Method for Multimodal Contrastive Learning

In recent years, data selection has emerged as a core issue for large-scale visual-language model pretraining, especially on noisy web-curated datasets. One widely adopted strategy assigns quality scores such as CLIP similarity for each sample and retains the data pairs with the highest scores. However, these approaches are agnostic of data distribution and always fail to select the most informative samples. To solve this problem, we propose a simple yet theoretically principled metric named Variance Alignment Score (VAS), which has the form $\langle \Sigma_{\text{test}}, \Sigma_i\rangle$. Here, $\Sigma_{\text{test}}$ represents the target (cross-)covariance matrix we aim to align, potentially based on prior knowledge, while $\Sigma_i$ denotes the tensor product of single or multi-modal representations for the $i$-th sample. We further design a new data selection method that maximizes the total VAS. We provide theoretical analysis in a simplified setting to demonstrate the theoretical advantage of VAS over random or other existing data selection. Experimentally, applying VAS and CLIP scores together can outperform baselines by a margin of $1.3\%$ average on 38 evaluation sets for noisy dataset DataComp and $2.5\%$ on VTAB for high-quality dataset CC12M. Additionally, our ablation study also shows visual features are better than text for calculating VAS, and the related classical experimental design methods may fail under this context.

An Experimental Design Framework for Label-Efficient Supervised Finetuning of Large Language Models

Supervised finetuning (SFT) on instruction datasets has played a crucial role in achieving the remarkable zero-shot generalization capabilities observed in modern large language models (LLMs). However, the annotation efforts required to produce high quality responses for instructions are becoming prohibitively expensive, especially as the number of tasks spanned by instruction datasets continues to increase. Active learning is effective in identifying useful subsets of samples to annotate from an unlabeled pool, but its high computational cost remains a barrier to its widespread applicability in the context of LLMs. To mitigate the annotation cost of SFT and circumvent the computational bottlenecks of active learning, we propose using experimental design. Experimental design techniques select the most informative samples to label, and typically maximize some notion of uncertainty and/or diversity. In our work, we implement a framework that evaluates several existing and novel experimental design techniques and find that these methods consistently yield significant gains in label efficiency with little computational overhead. On generative tasks, our methods achieve the same generalization performance with only $50\%$ of annotation cost required by random sampling.

Fair Active Learning in Low-Data Regimes

In critical machine learning applications, ensuring fairness is essential to avoid perpetuating social inequities. In this work, we address the challenges of reducing bias and improving accuracy in data-scarce environments, where the cost of collecting labeled data prohibits the use of large, labeled datasets. In such settings, active learning promises to maximize marginal accuracy gains of small amounts of labeled data. However, existing applications of active learning for fairness fail to deliver on this, typically requiring large labeled datasets, or failing to ensure the desired fairness tolerance is met on the population distribution. To address such limitations, we introduce an innovative active learning framework that combines an exploration procedure inspired by posterior sampling with a fair classification subroutine. We demonstrate that this framework performs effectively in very data-scarce regimes, maximizing accuracy while satisfying fairness constraints with high probability. We evaluate our proposed approach using well-established real-world benchmark datasets and compare it against state-of-the-art methods, demonstrating its effectiveness in producing fair models, and improvement over existing methods.

Minimax Optimal Submodular Optimization with Bandit Feedback

We consider maximizing a monotonic, submodular set function $f: 2^{[n]} \rightarrow [0,1]$ under stochastic bandit feedback. Specifically, $f$ is unknown to the learner but at each time $t=1,\dots,T$ the learner chooses a set $S_t \subset [n]$ with $|S_t| \leq k$ and receives reward $f(S_t) + \eta_t$ where $\eta_t$ is mean-zero sub-Gaussian noise. The objective is to minimize the learner's regret over $T$ times with respect to ($1-e^{-1}$)-approximation of maximum $f(S_*)$ with $|S_*| = k$, obtained through greedy maximization of $f$. To date, the best regret bound in the literature scales as $k n^{1/3} T^{2/3}$. And by trivially treating every set as a unique arm one deduces that $\sqrt{ {n \choose k} T }$ is also achievable. In this work, we establish the first minimax lower bound for this setting that scales like $\mathcal{O}(\min_{i \le k}(in^{1/3}T^{2/3} + \sqrt{n^{k-i}T}))$. Moreover, we propose an algorithm that is capable of matching the lower bound regret.

Near-Optimal Pure Exploration in Matrix Games: A Generalization of Stochastic Bandits & Dueling Bandits

We study the sample complexity of identifying the pure strategy Nash equilibrium (PSNE) in a two-player zero-sum matrix game with noise. Formally, we are given a stochastic model where any learner can sample an entry $(i,j)$ of the input matrix $A\in[-1,1]^{n\times m}$ and observe $A_{i,j}+\eta$ where $\eta$ is a zero-mean 1-sub-Gaussian noise. The aim of the learner is to identify the PSNE of $A$, whenever it exists, with high probability while taking as few samples as possible. Zhou et al. (2017) presents an instance-dependent sample complexity lower bound that depends only on the entries in the row and column in which the PSNE lies. We design a near-optimal algorithm whose sample complexity matches the lower bound, up to log factors. The problem of identifying the PSNE also generalizes the problem of pure exploration in stochastic multi-armed bandits and dueling bandits, and our result matches the optimal bounds, up to log factors, in both the settings.