Yifei Ma

Anusha Lalitha, Kousha Kalantari, Yifei Ma et al.

We study the problem of best-arm identification (BAI) in the fixed-budget setting with heterogeneous reward variances. We propose two variance-adaptive BAI algorithms for this setting: SHVar for known reward variances and SHAdaVar for unknown reward variances. Our algorithms rely on non-uniform budget allocations among the arms where the arms with higher reward variances are pulled more often than those with lower variances. The main algorithmic novelty is in the design of SHAdaVar, which allocates budget greedily based on overestimating the unknown reward variances. We bound probabilities of misidentifying the best arms in both SHVar and SHAdaVar. Our analyses rely on novel lower bounds on the number of pulls of an arm that do not require closed-form solutions to the budget allocation problem. Since one of our budget allocation problems is analogous to the optimal experiment design with unknown variances, we believe that our results are of a broad interest. Our experiments validate our theory, and show that SHVar and SHAdaVar outperform algorithms from prior works with analytical guarantees.

Haoyang Fang, Shuai Zhang, Yifei Ma et al.

Domain-specific finetuning is essential for dense retrievers, yet not all training pairs contribute equally to the learning process. We introduce OPERA, a data pruning framework that exploits this heterogeneity to improve both the effectiveness and efficiency of retrieval model adaptation. We first investigate static pruning (SP), which retains only high-similarity query-document pairs, revealing an intrinsic quality-coverage tradeoff: ranking (NDCG) improves while retrieval (Recall) can degrade due to reduced query diversity. To resolve this tradeoff, we propose a two-stage dynamic pruning (DP) strategy that adaptively modulates sampling probabilities at both query and document levels throughout training, prioritizing high-quality examples while maintaining access to the full training set. Evaluations across eight datasets spanning six domains demonstrate the effectiveness of both approaches: SP improves ranking over standard finetuning (NDCG@10 +0.5\%), while DP achieves the strongest performance on both ranking (NDCG@10 +1.9\%) and retrieval (Recall@20 +0.7\%), with an average rank of 1.38 across all methods. These findings scale to Qwen3-Embedding, an LLM-based dense retriever, confirming architecture-agnostic benefits. Notably, DP reaches comparable performance in less than 50\% of the training time required by standard finetuning.

Qiming He, Jing Li, Tian Guan et al.

Glomerular pathology is central to the diagnosis and prognosis of renal diseases, yet the heterogeneity of glomerular morphology and fine-grained lesion patterns remain challenging for current AI approaches. We present GloPath, an entity-centric foundation model trained on over one million glomeruli extracted from 14,049 renal biopsy specimens using multi-scale and multi-view self-supervised learning. GloPath addresses two major challenges in nephropathology: glomerular lesion assessment and clinicopathological insights discovery. For lesion assessment, GloPath was benchmarked across three independent cohorts on 52 tasks, including lesion recognition, grading, few-shot classification, and cross-modality diagnosis-outperforming state-of-the-art methods in 42 tasks (80.8%). In the large-scale real-world study, it achieved an ROC-AUC of 91.51% for lesion recognition, demonstrating strong robustness in routine clinical settings. For clinicopathological insights, GloPath systematically revealed statistically significant associations between glomerular morphological parameters and clinical indicators across 224 morphology-clinical variable pairs, demonstrating its capacity to connect tissue-level pathology with patient-level outcomes. Together, these results position GloPath as a scalable and interpretable platform for glomerular lesion assessment and clinicopathological discovery, representing a step toward clinically translatable AI in renal pathology.

Yaxuan Li, Han Jiang, Yifei Ma et al.

Dynamic Magnetic Resonance Imaging (MRI) of the vocal tract has become an increasingly adopted imaging modality for speech motor studies. Beyond image signals, systematic data loss, noise pollution, and audio file corruption can occur due to the unpredictability of the MRI acquisition environment. In such cases, generating audio from images is critical for data recovery in both clinical and research applications. However, this remains challenging due to hardware constraints, acoustic interference, and data corruption. Existing solutions, such as denoising and multi-stage synthesis methods, face limitations in audio fidelity and generalizability. To address these challenges, we propose a Knowledge Enhanced Conditional Variational Autoencoder (KE-CVAE), a novel two-step "knowledge enhancement + variational inference" framework for generating speech audio signals from cine dynamic MRI sequences. This approach introduces two key innovations: (1) integration of unlabeled MRI data for knowledge enhancement, and (2) a variational inference architecture to improve generative modeling capacity. To the best of our knowledge, this is one of the first attempts at synthesizing speech audio directly from dynamic MRI video sequences. The proposed method was trained and evaluated on an open-source dynamic vocal tract MRI dataset recorded during speech. Experimental results demonstrate its effectiveness in generating natural speech waveforms while addressing MRI-specific acoustic challenges, outperforming conventional deep learning-based synthesis approaches.

Subhojyoti Mukherjee, Anusha Lalitha, Kousha Kalantari et al.

Learning of preference models from human feedback has been central to recent advances in artificial intelligence. Motivated by the cost of obtaining high-quality human annotations, we study efficient human preference elicitation for learning preference models. The key idea in our work is to generalize optimal designs, a methodology for computing optimal information-gathering policies, to questions with multiple answers, represented as lists of items. The policy is a distribution over lists and we elicit preferences from the list proportionally to its probability. To show the generality of our ideas, we study both absolute and ranking feedback models on items in the list. We design efficient algorithms for both and analyze them. Finally, we demonstrate that our algorithms are practical by evaluating them on existing question-answering problems.

Behnam Rahdari, Hao Ding, Ziwei Fan et al.

The unique capabilities of Large Language Models (LLMs), such as the natural language text generation ability, position them as strong candidates for providing explanation for recommendations. However, despite the size of the LLM, most existing models struggle to produce zero-shot explanations reliably. To address this issue, we propose a framework called Logic-Scaffolding, that combines the ideas of aspect-based explanation and chain-of-thought prompting to generate explanations through intermediate reasoning steps. In this paper, we share our experience in building the framework and present an interactive demonstration for exploring our results.

Subhojyoti Mukherjee, Anusha Lalitha, Aniket Deshmukh et al.

One emergent ability of large language models (LLMs) is that query-specific examples can be included in the prompt at inference time. In this work, we use active learning for adaptive prompt design and call it Active In-context Prompt Design (AIPD). We design the LLM prompt by adaptively choosing few-shot examples from a training set to optimize performance on a test set. The training examples are initially unlabeled and we obtain the label of the most informative ones, which maximally reduces uncertainty in the LLM prediction. We propose two algorithms, GO and SAL, which differ in how the few-shot examples are chosen. We analyze these algorithms in linear models: first GO and then use its equivalence with SAL. We experiment with many different tasks in small, medium-sized, and large language models; and show that GO and SAL outperform other methods for choosing few-shot examples in the LLM prompt at inference time.

Xinhao Liu, Jiaqi Li, Youming Deng et al.

Reproducible closed-loop evaluation remains a major bottleneck in Embodied AI such as visual navigation. A promising path forward is high-fidelity simulation that combines photorealistic sensor rendering with geometrically grounded interaction in complex, open-world urban environments. Although recent video-3DGS methods ease open-world scene capturing, they are still unsuitable for benchmarking due to large visual and geometric sim-to-real gaps. To address these challenges, we introduce Wanderland, a real-to-sim framework that features multi-sensor capture, reliable reconstruction, accurate geometry, and robust view synthesis. Using this pipeline, we curate a diverse dataset of indoor-outdoor urban scenes and systematically demonstrate how image-only pipelines scale poorly, how geometry quality impacts novel view synthesis, and how all of these adversely affect navigation policy learning and evaluation reliability. Beyond serving as a trusted testbed for embodied navigation, Wanderland's rich raw sensor data further allows benchmarking of 3D reconstruction and novel view synthesis models. Our work establishes a new foundation for reproducible research in open-world embodied AI. Project website is at https://ai4ce.github.io/wanderland/.

Yuanhao Li, Mingshan Liu, Hongbo Wang et al.

Large Language Models (LLMs) have shown impressive capabilities in multi-step reasoning and problem-solving.Recent works introduce multi-agent reflection frameworks where multiple LLM agents critique and refine each other's outputs using reinforcement learning (RL). However, these approaches often rely on single-shot responses and lack structural diversity in reasoning exploration. In this paper, we propose DRAFT-RL, a novel framework that integrates Chain-of-Draft (CoD) reasoning into multi-agent RL training. Instead of generating single responses, each agent produces multiple drafts per query, which are then evaluated by peer agents and a learned reward model to identify the most promising trajectory. These selected drafts are used to refine future reasoning strategies through actor-critic learning.DRAFT-RL enables explicit multi-path exploration, peer-guided reflection, and reward-aligned selection, resulting in more robust and interpretable LLM agent behavior. We evaluate our method on complex reasoning tasks including code synthesis, symbolic math, and knowledge-intensive QA,demonstrating that DRAFT-RL outperforms existing reflective and RL-based agents by significant margins in both accuracy and convergence speed

Jiawen Li, Tian Guan, Qingxin Xia et al.

Foundation models have revolutionized the paradigm of digital pathology, as they leverage general-purpose features to emulate real-world pathological practices, enabling the quantitative analysis of critical histological patterns and the dissection of cancer-specific signals. However, these static general features constrain the flexibility and pathological relevance in the ever-evolving needs of clinical applications, hindering the broad use of the current models. Here we introduce PathFiT, a dynamic feature learning method that can be effortlessly plugged into various pathology foundation models to unlock their adaptability. Meanwhile, PathFiT performs seamless implementation across diverse pathology applications regardless of downstream specificity. To validate PathFiT, we construct a digital pathology benchmark with over 20 terabytes of Internet and real-world data comprising 28 H\&E-stained tasks and 7 specialized imaging tasks including Masson's Trichrome staining and immunofluorescence images. By applying PathFiT to the representative pathology foundation models, we demonstrate state-of-the-art performance on 34 out of 35 tasks, with significant improvements on 23 tasks and outperforming by 10.20% on specialized imaging tasks. The superior performance and versatility of PathFiT open up new avenues in computational pathology.

Miaosen Zhang, Yixuan Wei, Zhen Xing et al.

Modern vision models are trained on very large noisy datasets. While these models acquire strong capabilities, they may not follow the user's intent to output the desired results in certain aspects, e.g., visual aesthetic, preferred style, and responsibility. In this paper, we target the realm of visual aesthetics and aim to align vision models with human aesthetic standards in a retrieval system. Advanced retrieval systems usually adopt a cascade of aesthetic models as re-rankers or filters, which are limited to low-level features like saturation and perform poorly when stylistic, cultural or knowledge contexts are involved. We find that utilizing the reasoning ability of large language models (LLMs) to rephrase the search query and extend the aesthetic expectations can make up for this shortcoming. Based on the above findings, we propose a preference-based reinforcement learning method that fine-tunes the vision models to distill the knowledge from both LLMs reasoning and the aesthetic models to better align the vision models with human aesthetics. Meanwhile, with rare benchmarks designed for evaluating retrieval systems, we leverage large multi-modality model (LMM) to evaluate the aesthetic performance with their strong abilities. As aesthetic assessment is one of the most subjective tasks, to validate the robustness of LMM, we further propose a novel dataset named HPIR to benchmark the alignment with human aesthetics. Experiments demonstrate that our method significantly enhances the aesthetic behaviors of the vision models, under several metrics. We believe the proposed algorithm can be a general practice for aligning vision models with human values.

Hao Wang, Yifei Ma, Hao Ding et al.

Recurrent neural networks have proven effective in modeling sequential user feedbacks for recommender systems. However, they usually focus solely on item relevance and fail to effectively explore diverse items for users, therefore harming the system performance in the long run. To address this problem, we propose a new type of recurrent neural networks, dubbed recurrent exploration networks (REN), to jointly perform representation learning and effective exploration in the latent space. REN tries to balance relevance and exploration while taking into account the uncertainty in the representations. Our theoretical analysis shows that REN can preserve the rate-optimal sublinear regret even when there exists uncertainty in the learned representations. Our empirical study demonstrates that REN can achieve satisfactory long-term rewards on both synthetic and real-world recommendation datasets, outperforming state-of-the-art models.

Hao Ding, Yifei Ma, Anoop Deoras et al.

Performance of recommender systems (RS) relies heavily on the amount of training data available. This poses a chicken-and-egg problem for early-stage products, whose amount of data, in turn, relies on the performance of their RS. On the other hand, zero-shot learning promises some degree of generalization from an old dataset to an entirely new dataset. In this paper, we explore the possibility of zero-shot learning in RS. We develop an algorithm, dubbed ZEro-Shot Recommenders (ZESRec), that is trained on an old dataset and generalize to a new one where there are neither overlapping users nor overlapping items, a setting that contrasts typical cross-domain RS that has either overlapping users or items. Different from categorical item indices, i.e., item ID, in previous methods, ZESRec uses items' natural-language descriptions (or description embeddings) as their continuous indices, and therefore naturally generalize to any unseen items. In terms of users, ZESRec builds upon recent advances on sequential RS to represent users using their interactions with items, thereby generalizing to unseen users as well. We study three pairs of real-world RS datasets and demonstrate that ZESRec can successfully enable recommendations in such a zero-shot setting, opening up new opportunities for resolving the chicken-and-egg problem for data-scarce startups or early-stage products.

Tengyang Xie, Yifei Ma, Yu-Xiang Wang

Motivated by the many real-world applications of reinforcement learning (RL) that require safe-policy iterations, we consider the problem of off-policy evaluation (OPE) -- the problem of evaluating a new policy using the historical data obtained by different behavior policies -- under the model of nonstationary episodic Markov Decision Processes (MDP) with a long horizon and a large action space. Existing importance sampling (IS) methods often suffer from large variance that depends exponentially on the RL horizon $H$. To solve this problem, we consider a marginalized importance sampling (MIS) estimator that recursively estimates the state marginal distribution for the target policy at every step. MIS achieves a mean-squared error of $$ \frac{1}{n} \sum\nolimits_{t=1}^H\mathbb{E}_μ\left[\frac{d_t^π(s_t)^2}{d_t^μ(s_t)^2} \mathrm{Var}_μ\left[\frac{π_t(a_t|s_t)}{μ_t(a_t|s_t)}\big( V_{t+1}^π(s_{t+1}) + r_t\big) \middle| s_t\right]\right] + \tilde{O}(n^{-1.5}) $$ where $μ$ and $π$ are the logging and target policies, $d_t^μ(s_t)$ and $d_t^π(s_t)$ are the marginal distribution of the state at $t$th step, $H$ is the horizon, $n$ is the sample size and $V_{t+1}^π$ is the value function of the MDP under $π$. The result matches the Cramer-Rao lower bound in \citet{jiang2016doubly} up to a multiplicative factor of $H$. To the best of our knowledge, this is the first OPE estimation error bound with a polynomial dependence on $H$. Besides theory, we show empirical superiority of our method in time-varying, partially observable, and long-horizon RL environments.

Haibin Lin, Hang Zhang, Yifei Ma et al.

With an increasing demand for training powers for deep learning algorithms and the rapid growth of computation resources in data centers, it is desirable to dynamically schedule different distributed deep learning tasks to maximize resource utilization and reduce cost. In this process, different tasks may receive varying numbers of machines at different time, a setting we call elastic distributed training. Despite the recent successes in large mini-batch distributed training, these methods are rarely tested in elastic distributed training environments and suffer degraded performance in our experiments, when we adjust the learning rate linearly immediately with respect to the batch size. One difficulty we observe is that the noise in the stochastic momentum estimation is accumulated over time and will have delayed effects when the batch size changes. We therefore propose to smoothly adjust the learning rate over time to alleviate the influence of the noisy momentum estimation. Our experiments on image classification, object detection and semantic segmentation have demonstrated that our proposed Dynamic SGD method achieves stabilized performance when varying the number of GPUs from 8 to 128. We also provide theoretical understanding on the optimality of linear learning rate scheduling and the effects of stochastic momentum.

Yifei Ma, Yu-Xiang Wang, Balakrishnan et al.

We study the problem of offline learning in automated decision systems under the contextual bandits model. We are given logged historical data consisting of contexts, (randomized) actions, and (nonnegative) rewards. A common goal is to evaluate what would happen if different actions were taken in the same contexts, so as to optimize the action policies accordingly. The typical approach to this problem, inverse probability weighted estimation (IPWE) [Bottou et al., 2013], requires logged action probabilities, which may be missing in practice due to engineering complications. Even when available, small action probabilities cause large uncertainty in IPWE, rendering the corresponding results insignificant. To solve both problems, we show how one can use policy improvement (PIL) objectives, regularized by policy imitation (IML). We motivate and analyze PIL as an extension to Clipped-IPWE, by showing that both are lower-bound surrogates to the vanilla IPWE. We also formally connect IML to IPWE variance estimation [Swaminathan and Joachims 2015] and natural policy gradients. Without probability logging, our PIL-IML interpretations justify and improve, by reward-weighting, the state-of-art cross-entropy (CE) loss that predicts the action items among all action candidates available in the same contexts. With probability logging, our main theoretical contribution connects IML-underfitting to the existence of either confounding variables or model misspecification. We show the value and accuracy of our insights by simulations based on Simpson's paradox, standard UCI multiclass-to-bandit conversions and on the Criteo counterfactual analysis challenge dataset.

Hadi Salman, Elif Ayvali, Rangaprasad Arun Srivatsan et al.

In this work, we develop an approach for guiding robots to automatically localize and find the shapes of tumors and other stiff inclusions present in the anatomy. Our approach uses Gaussian processes to model the stiffness distribution and active learning to direct the palpation path of the robot. The palpation paths are chosen such that they maximize an acquisition function provided by an active learning algorithm. Our approach provides the flexibility to avoid obstacles in the robot's path, incorporate uncertainties in robot position and sensor measurements, include prior information about location of stiff inclusions while respecting the robot-kinematics. To the best of our knowledge this is the first work in literature that considers all the above conditions while localizing tumors. The proposed framework is evaluated via simulation and experimentation on three different robot platforms: 6-DoF industrial arm, da Vinci Research Kit (dVRK), and the Insertable Robotic Effector Platform (IREP). Results show that our approach can accurately estimate the locations and boundaries of the stiff inclusions while reducing exploration time.

Yifei Ma, Roman Garnett, Jeff Schneider

Autonomous systems can be used to search for sparse signals in a large space; e.g., aerial robots can be deployed to localize threats, detect gas leaks, or respond to distress calls. Intuitively, search algorithms may increase efficiency by collecting aggregate measurements summarizing large contiguous regions. However, most existing search methods either ignore the possibility of such region observations (e.g., Bayesian optimization and multi-armed bandits) or make strong assumptions about the sensing mechanism that allow each measurement to arbitrarily encode all signals in the entire environment (e.g., compressive sensing). We propose an algorithm that actively collects data to search for sparse signals using only noisy measurements of the average values on rectangular regions (including single points), based on the greedy maximization of information gain. We analyze our algorithm in 1d and show that it requires $\tilde{O}(\frac{n}{μ^2}+k^2)$ measurements to recover all of $k$ signal locations with small Bayes error, where $μ$ and $n$ are the signal strength and the size of the search space, respectively. We also show that active designs can be fundamentally more efficient than passive designs with region sensing, contrasting with the results of Arias-Castro, Candes, and Davenport (2013). We demonstrate the empirical performance of our algorithm on a search problem using satellite image data and in high dimensions.

Yifei Ma, Roman Garnett, Jeff Schneider

Many real-world datasets can be represented in the form of a graph whose edge weights designate similarities between instances. A discrete Gaussian random field (GRF) model is a finite-dimensional Gaussian process (GP) whose prior covariance is the inverse of a graph Laplacian. Minimizing the trace of the predictive covariance Sigma (V-optimality) on GRFs has proven successful in batch active learning classification problems with budget constraints. However, its worst-case bound has been missing. We show that the V-optimality on GRFs as a function of the batch query set is submodular and hence its greedy selection algorithm guarantees an (1-1/e) approximation ratio. Moreover, GRF models have the absence-of-suppressor (AofS) condition. For active survey problems, we propose a similar survey criterion which minimizes 1'(Sigma)1. In practice, V-optimality criterion performs better than GPs with mutual information gain criteria and allows nonuniform costs for different nodes.