Haohan Wang

Zeyi Huang, Haohan Wang, Dong Huang et al. · cmu

Training with an emphasis on "hard-to-learn" components of the data has been proven as an effective method to improve the generalization of machine learning models, especially in the settings where robustness (e.g., generalization across distributions) is valued. Existing literature discussing this "hard-to-learn" concept are mainly expanded either along the dimension of the samples or the dimension of the features. In this paper, we aim to introduce a simple view merging these two dimensions, leading to a new, simple yet effective, heuristic to train machine learning models by emphasizing the worst-cases on both the sample and the feature dimensions. We name our method W2D following the concept of "Worst-case along Two Dimensions". We validate the idea and demonstrate its empirical strength over standard benchmarks.

Andy Zhou, Kai Yan, Michal Shlapentokh-Rothman et al.

While language models (LMs) have shown potential across a range of decision-making tasks, their reliance on simple acting processes limits their broad deployment as autonomous agents. In this paper, we introduce Language Agent Tree Search (LATS) -- the first general framework that synergizes the capabilities of LMs in reasoning, acting, and planning. By leveraging the in-context learning ability of LMs, we integrate Monte Carlo Tree Search into LATS to enable LMs as agents, along with LM-powered value functions and self-reflections for proficient exploration and enhanced decision-making. A key feature of our approach is the incorporation of an environment for external feedback, which offers a more deliberate and adaptive problem-solving mechanism that surpasses the constraints of existing techniques. Our experimental evaluation across diverse domains, including programming, interactive question-answering (QA), web navigation, and math, validates the effectiveness and generality of LATS in decision-making while maintaining competitive or improved reasoning performance. Notably, LATS achieves state-of-the-art pass@1 accuracy (92.7%) for programming on HumanEval with GPT-4 and demonstrates gradient-free performance (average score of 75.9) comparable to gradient-based fine-tuning for web navigation on WebShop with GPT-3.5. Code can be found at https://github.com/lapisrocks/LanguageAgentTreeSearch

Chonghan Chen, Haohan Wang, Leyang Hu et al.

We introduce the initial release of our software Robustar, which aims to improve the robustness of vision classification machine learning models through a data-driven perspective. Building upon the recent understanding that the lack of machine learning model's robustness is the tendency of the model's learning of spurious features, we aim to solve this problem from its root at the data perspective by removing the spurious features from the data before training. In particular, we introduce a software that helps the users to better prepare the data for training image classification models by allowing the users to annotate the spurious features at the pixel level of images. To facilitate this process, our software also leverages recent advances to help identify potential images and pixels worthy of attention and to continue the training with newly annotated data. Our software is hosted at the GitHub Repository https://github.com/HaohanWang/Robustar.

Minh-Long Luu, Zeyi Huang, Eric P. Xing et al.

Mix-up training approaches have proven to be effective in improving the generalization ability of Deep Neural Networks. Over the years, the research community expands mix-up methods into two directions, with extensive efforts to improve saliency-guided procedures but minimal focus on the arbitrary path, leaving the randomization domain unexplored. In this paper, inspired by the superior qualities of each direction over one another, we introduce a novel method that lies at the junction of the two routes. By combining the best elements of randomness and saliency utilization, our method balances speed, simplicity, and accuracy. We name our method R-Mix following the concept of "Random Mix-up". We demonstrate its effectiveness in generalization, weakly supervised object localization, calibration, and robustness to adversarial attacks. Finally, in order to address the question of whether there exists a better decision protocol, we train a Reinforcement Learning agent that decides the mix-up policies based on the classifier's performance, reducing dependency on human-designed objectives and hyperparameter tuning. Extensive experiments further show that the agent is capable of performing at the cutting-edge level, laying the foundation for a fully automatic mix-up. Our code is released at [https://github.com/minhlong94/Random-Mixup].

Kun Xiang, Xing Zhang, Jinwen She et al.

As the COVID-19 pandemic puts pressure on healthcare systems worldwide, the computed tomography image based AI diagnostic system has become a sustainable solution for early diagnosis. However, the model-wise vulnerability under adversarial perturbation hinders its deployment in practical situation. The existing adversarial training strategies are difficult to generalized into medical imaging field challenged by complex medical texture features. To overcome this challenge, we propose a Contour Attention Preserving (CAP) method based on lung cavity edge extraction. The contour prior features are injected to attention layer via a parameter regularization and we optimize the robust empirical risk with hybrid distance metric. We then introduce a new cross-nation CT scan dataset to evaluate the generalization capability of the adversarial robustness under distribution shift. Experimental results indicate that the proposed method achieves state-of-the-art performance in multiple adversarial defense and generalization tasks. The code and dataset are available at https://github.com/Quinn777/CAP.

Yifan Zhong, Haohan Wang, Eric P. Xing

Many recent neural models have shown remarkable empirical results in Machine Reading Comprehension, but evidence suggests sometimes the models take advantage of dataset biases to predict and fail to generalize on out-of-sample data. While many other approaches have been proposed to address this issue from the computation perspective such as new architectures or training procedures, we believe a method that allows researchers to discover biases, and adjust the data or the models in an earlier stage will be beneficial. Thus, we introduce MRCLens, a toolkit that detects whether biases exist before users train the full model. For the convenience of introducing the toolkit, we also provide a categorization of common biases in MRC.

Peiyan Zhang, Haoyang Liu, Chaozhuo Li et al.

Machine learning has demonstrated remarkable performance over finite datasets, yet whether the scores over the fixed benchmarks can sufficiently indicate the model's performance in the real world is still in discussion. In reality, an ideal robust model will probably behave similarly to the oracle (e.g., the human users), thus a good evaluation protocol is probably to evaluate the models' behaviors in comparison to the oracle. In this paper, we introduce a new robustness measurement that directly measures the image classification model's performance compared with a surrogate oracle (i.e., a foundation model). Besides, we design a simple method that can accomplish the evaluation beyond the scope of the benchmarks. Our method extends the image datasets with new samples that are sufficiently perturbed to be distinct from the ones in the original sets, but are still bounded within the same image-label structure the original test image represents, constrained by a foundation model pretrained with a large amount of samples. As a result, our new method will offer us a new way to evaluate the models' robustness performance, free of limitations of fixed benchmarks or constrained perturbations, although scoped by the power of the oracle. In addition to the evaluation results, we also leverage our generated data to understand the behaviors of the model and our new evaluation strategies.

Esla Timothy Anzaku, Haohan Wang, Arnout Van Messem et al.

Deep Neural Network (DNN) models are increasingly evaluated using new replication test datasets, which have been carefully created to be similar to older and popular benchmark datasets. However, running counter to expectations, DNN classification models show significant, consistent, and largely unexplained degradation in accuracy on these replication test datasets. While the popular evaluation approach is to assess the accuracy of a model by making use of all the datapoints available in the respective test datasets, we argue that doing so hinders us from adequately capturing the behavior of DNN models and from having realistic expectations about their accuracy. Therefore, we propose a principled evaluation protocol that is suitable for performing comparative investigations of the accuracy of a DNN model on multiple test datasets, leveraging subsets of datapoints that can be selected using different criteria, including uncertainty-related information. By making use of this new evaluation protocol, we determined the accuracy of $564$ DNN models on both (1) the CIFAR-10 and ImageNet datasets and (2) their replication datasets. Our experimental results indicate that the observed accuracy degradation between established benchmark datasets and their replications is consistently lower (that is, models do perform better on the replication test datasets) than the accuracy degradation reported in published works, with these published works relying on conventional evaluation approaches that do not utilize uncertainty-related information.

Wang Lu, Hao Yu, Jindong Wang et al.

When personalized federated learning (FL) meets large foundation models, new challenges arise from various limitations in resources. In addition to typical limitations such as data, computation, and communication costs, access to the models is also often limited. This paper endeavors to solve both the challenges of limited resources and personalization. i.e., distribution shifts between clients. To do so, we propose a method named ZOOPFL that uses Zeroth-Order Optimization for Personalized Federated Learning. ZOOPFL avoids direct interference with the foundation models and instead learns to adapt its inputs through zeroth-order optimization. In addition, we employ simple yet effective linear projections to remap its predictions for personalization. To reduce the computation costs and enhance personalization, we propose input surgery to incorporate an auto-encoder with low-dimensional and client-specific embeddings. We provide theoretical support for ZOOPFL to analyze its convergence. Extensive empirical experiments on computer vision and natural language processing tasks using popular foundation models demonstrate its effectiveness for FL on black-box foundation models.

Ke Chen, Yifeng Wang, Yufei Zhou et al. · cmu

In the field of Alzheimer's disease diagnosis, segmentation and classification tasks are inherently interconnected. Sharing knowledge between models for these tasks can significantly improve training efficiency, particularly when training data is scarce. However, traditional knowledge distillation techniques often struggle to bridge the gap between segmentation and classification due to the distinct nature of tasks and different model architectures. To address this challenge, we propose a dual-stream pipeline that facilitates cross-task and cross-architecture knowledge sharing. Our approach introduces a dual-stream embedding module that unifies feature representations from segmentation and classification models, enabling dimensional integration of these features to guide the classification model. We validated our method on multiple 3D datasets for Alzheimer's disease diagnosis, demonstrating significant improvements in classification performance, especially on small datasets. Furthermore, we extended our pipeline with a residual temporal attention mechanism for early diagnosis, utilizing images taken before the atrophy of patients' brain mass. This advancement shows promise in enabling diagnosis approximately six months earlier in mild and asymptomatic stages, offering critical time for intervention.

Thomas Yu Chow Tam, Litian Liang, Ke Chen et al.

Deep learning (DL) models have shown significant potential in Alzheimer's Disease (AD) classification. However, understanding and interpreting these models remains challenging, which hinders the adoption of these models in clinical practice. Techniques such as saliency maps have been proven effective in providing visual and empirical clues about how these models work, but there still remains a gap in understanding which specific brain regions DL models focus on and whether these brain regions are pathologically associated with AD. To bridge such gap, in this study, we developed a quantitative disease-focusing strategy to first enhance the interpretability of DL models using saliency maps and brain segmentations; then we propose a disease-focus (DF) score that quantifies how much a DL model focuses on brain areas relevant to AD pathology based on clinically known MRI-based pathological regions of AD. Using this strategy, we compared several state-of-the-art DL models, including a baseline 3D ResNet model, a pretrained MedicalNet model, and a MedicalNet with data augmentation to classify patients with AD vs. cognitive normal patients using MRI data; then we evaluated these models in terms of their abilities to focus on disease-relevant regions. Our results show interesting disease-focusing patterns with different models, particularly characteristic patterns with the pretrained models and data augmentation, and also provide insight into their classification performance. These results suggest that the approach we developed for quantitatively assessing the abilities of DL models to focus on disease-relevant regions may help improve interpretability of these models for AD classification and facilitate their adoption for AD diagnosis in clinical practice. The code is publicly available at https://github.com/Liang-lt/ADNI.

Andy Zhou, Jindong Wang, Yu-Xiong Wang et al.

We propose a conceptually simple and lightweight framework for improving the robustness of vision models through the combination of knowledge distillation and data augmentation. We address the conjecture that larger models do not make for better teachers by showing strong gains in out-of-distribution robustness when distilling from pretrained foundation models. Following this finding, we propose Discrete Adversarial Distillation (DAD), which leverages a robust teacher to generate adversarial examples and a VQGAN to discretize them, creating more informative samples than standard data augmentation techniques. We provide a theoretical framework for the use of a robust teacher in the knowledge distillation with data augmentation setting and demonstrate strong gains in out-of-distribution robustness and clean accuracy across different student architectures. Notably, our method adds minor computational overhead compared to similar techniques and can be easily combined with other data augmentations for further improvements.

Haoyang Liu, Yijiang Li, Tiancheng Xing et al.

Dataset Distillation (DD) aims to generate a compact synthetic dataset that enables models to achieve performance comparable to training on the full large dataset, significantly reducing computational costs. Drawing from optimal transport theory, we introduce WMDD (Wasserstein Metric-based Dataset Distillation), a straightforward yet powerful method that employs the Wasserstein metric to enhance distribution matching. We compute the Wasserstein barycenter of features from a pretrained classifier to capture essential characteristics of the original data distribution. By optimizing synthetic data to align with this barycenter in feature space and leveraging per-class BatchNorm statistics to preserve intra-class variations, WMDD maintains the efficiency of distribution matching approaches while achieving state-of-the-art results across various high-resolution datasets. Our extensive experiments demonstrate WMDD's effectiveness and adaptability, highlighting its potential for advancing machine learning applications at scale.

Eric Xue, Yijiang Li, Haoyang Liu et al.

Dataset distillation (DD) allows datasets to be distilled to fractions of their original size while preserving the rich distributional information, so that models trained on the distilled datasets can achieve a comparable accuracy while saving significant computational loads. Recent research in this area has been focusing on improving the accuracy of models trained on distilled datasets. In this paper, we aim to explore a new perspective of DD. We study how to embed adversarial robustness in distilled datasets, so that models trained on these datasets maintain the high accuracy and meanwhile acquire better adversarial robustness. We propose a new method that achieves this goal by incorporating curvature regularization into the distillation process with much less computational overhead than standard adversarial training. Extensive empirical experiments suggest that our method not only outperforms standard adversarial training on both accuracy and robustness with less computation overhead but is also capable of generating robust distilled datasets that can withstand various adversarial attacks. Our implementation is available at: https://github.com/yumozi/GUARD.

Tong Zhang, Haoyang Liu, Peiyan Zhang et al.

In the field of computer graphics, the use of vector graphics, particularly Scalable Vector Graphics (SVG), represents a notable development from traditional pixel-based imagery. SVGs, with their XML-based format, are distinct in their ability to directly and explicitly represent visual elements such as shape, color, and path. This direct representation facilitates a more accurate and logical depiction of graphical elements, enhancing reasoning and interpretability. Recognizing the potential of SVGs, the machine learning community has introduced multiple methods for image vectorization. However, transforming images into SVG format while retaining the relational properties and context of the original scene remains a key challenge. Most vectorization methods often yield SVGs that are overly complex and not easily interpretable. In response to this challenge, we introduce our method, Simple-SVG-Generation (S\textsuperscript{2}VG\textsuperscript{2}). Our method focuses on producing SVGs that are both accurate and simple, aligning with human readability and understanding. With simple images, we evaluate our method with reasoning tasks together with advanced language models, the results show a clear improvement over previous SVG generation methods. We also conducted surveys for human evaluation on the readability of our generated SVGs, the results also favor our methods.

Yijiang Li, Ying Gao, Haohan Wang

We investigate a specific security risk in FL: a group of malicious clients has impacted the model during training by disguising their identities and acting as benign clients but later switching to an adversarial role. They use their data, which was part of the training set, to train a substitute model and conduct transferable adversarial attacks against the federated model. This type of attack is subtle and hard to detect because these clients initially appear to be benign. The key question we address is: How robust is the FL system to such covert attacks, especially compared to traditional centralized learning systems? We empirically show that the proposed attack imposes a high security risk to current FL systems. By using only 3\% of the client's data, we achieve the highest attack rate of over 80\%. To further offer a full understanding of the challenges the FL system faces in transferable attacks, we provide a comprehensive analysis over the transfer robustness of FL across a spectrum of configurations. Surprisingly, FL systems show a higher level of robustness than their centralized counterparts, especially when both systems are equally good at handling regular, non-malicious data. We attribute this increased robustness to two main factors: 1) Decentralized Data Training: Each client trains the model on its own data, reducing the overall impact of any single malicious client. 2) Model Update Averaging: The updates from each client are averaged together, further diluting any malicious alterations. Both practical experiments and theoretical analysis support our conclusions. This research not only sheds light on the resilience of FL systems against hidden attacks but also raises important considerations for their future application and development.

Haoyang Liu, Yijiang Li, Haohan Wang

Gene expression analysis holds the key to many biomedical discoveries, yet extracting insights from raw transcriptomic data remains formidable due to the complexity of multiple large, semi-structured files and the need for extensive domain expertise. Current automation approaches are often limited by either inflexible workflows that break down in edge cases or by fully autonomous agents that lack the necessary precision for rigorous scientific inquiry. GenoMAS charts a different course by presenting a team of LLM-based scientists that integrates the reliability of structured workflows with the adaptability of autonomous agents. GenoMAS orchestrates six specialized LLM agents through typed message-passing protocols, each contributing complementary strengths to a shared analytic canvas. At the heart of GenoMAS lies a guided-planning framework: programming agents unfold high-level task guidelines into Action Units and, at each juncture, elect to advance, revise, bypass, or backtrack, thereby maintaining logical coherence while bending gracefully to the idiosyncrasies of genomic data. On the GenoTEX benchmark, GenoMAS reaches a Composite Similarity Correlation of 89.13% for data preprocessing and an F$_1$ of 60.48% for gene identification, surpassing the best prior art by 10.61% and 16.85% respectively. Beyond metrics, GenoMAS surfaces biologically plausible gene-phenotype associations corroborated by the literature, all while adjusting for latent confounders. Code is available at https://github.com/Liu-Hy/GenoMAS.

Jingfeng Zhang, Bo Song, Haohan Wang et al.

Label-noise learning (LNL) aims to increase the model's generalization given training data with noisy labels. To facilitate practical LNL algorithms, researchers have proposed different label noise types, ranging from class-conditional to instance-dependent noises. In this paper, we introduce a novel label noise type called BadLabel, which can significantly degrade the performance of existing LNL algorithms by a large margin. BadLabel is crafted based on the label-flipping attack against standard classification, where specific samples are selected and their labels are flipped to other labels so that the loss values of clean and noisy labels become indistinguishable. To address the challenge posed by BadLabel, we further propose a robust LNL method that perturbs the labels in an adversarial manner at each epoch to make the loss values of clean and noisy labels again distinguishable. Once we select a small set of (mostly) clean labeled data, we can apply the techniques of semi-supervised learning to train the model accurately. Empirically, our experimental results demonstrate that existing LNL algorithms are vulnerable to the newly introduced BadLabel noise type, while our proposed robust LNL method can effectively improve the generalization performance of the model under various types of label noise. The new dataset of noisy labels and the source codes of robust LNL algorithms are available at https://github.com/zjfheart/BadLabels.

Bang Liu, Xinfeng Li, Jiayi Zhang et al. · microsoft-research

The advent of large language models (LLMs) has catalyzed a transformative shift in artificial intelligence, paving the way for advanced intelligent agents capable of sophisticated reasoning, robust perception, and versatile action across diverse domains. As these agents increasingly drive AI research and practical applications, their design, evaluation, and continuous improvement present intricate, multifaceted challenges. This book provides a comprehensive overview, framing intelligent agents within modular, brain-inspired architectures that integrate principles from cognitive science, neuroscience, and computational research. We structure our exploration into four interconnected parts. First, we systematically investigate the modular foundation of intelligent agents, systematically mapping their cognitive, perceptual, and operational modules onto analogous human brain functionalities and elucidating core components such as memory, world modeling, reward processing, goal, and emotion. Second, we discuss self-enhancement and adaptive evolution mechanisms, exploring how agents autonomously refine their capabilities, adapt to dynamic environments, and achieve continual learning through automated optimization paradigms. Third, we examine multi-agent systems, investigating the collective intelligence emerging from agent interactions, cooperation, and societal structures. Finally, we address the critical imperative of building safe and beneficial AI systems, emphasizing intrinsic and extrinsic security threats, ethical alignment, robustness, and practical mitigation strategies necessary for trustworthy real-world deployment. By synthesizing modular AI architectures with insights from different disciplines, this survey identifies key research challenges and opportunities, encouraging innovations that harmonize technological advancement with meaningful societal benefit.

Yaoning Yu, Kai-Min Chang, Ye Yu et al.

Financial documents like earning reports or balance sheets often involve long tables and multi-page reports. Large language models have become a new tool to help numerical reasoning and understanding these documents. However, prompt quality can have a major effect on how well LLMs perform these financial reasoning tasks. Most current methods tune prompts on fixed datasets of financial text or tabular data, which limits their ability to adapt to new question types or document structures, or they involve costly and manually labeled/curated dataset to help build the prompts. We introduce a self-improving prompt framework driven by data-augmented optimization. In this closed-loop process, we generate synthetic financial tables and document excerpts, verify their correctness and robustness, and then update the prompt based on the results. Specifically, our framework combines a synthetic data generator with verifiers and a prompt optimizer, where the generator produces new examples that exposes weaknesses in the current prompt, the verifiers check the validity and robustness of the produced examples, and the optimizer incrementally refines the prompt in response. By iterating these steps in a feedback cycle, our method steadily improves prompt accuracy on financial reasoning tasks without needing external labels. Evaluation on DocMath-Eval benchmark demonstrates that our system achieves higher performance in both accuracy and robustness than standard prompt methods, underscoring the value of incorporating synthetic data generation into prompt learning for financial applications.

Yihan Zhang, Xuanshuo Zhang, Wei Wu et al.

Saliency maps have been widely used to interpret deep learning classifiers for Alzheimer's disease (AD). However, since AD is heterogeneous and has multiple subtypes, the pathological mechanism of AD remains not fully understood and may vary from patient to patient. Due to the lack of such understanding, it is difficult to comprehensively and effectively assess the saliency map of AD classifier. In this paper, we utilize the anatomical segmentation to allocate saliency values into different brain regions. By plotting the distributions of saliency maps corresponding to AD and NC (Normal Control), we can gain a comprehensive view of the model's decisions process. In order to leverage the fact that the brain volume shrinkage happens in AD patients during disease progression, we define a new evaluation metric, brain volume change score (VCS), by computing the average Pearson correlation of the brain volume changes and the saliency values of a model in different brain regions for each patient. Thus, the VCS metric can help us gain some knowledge of how saliency maps resulting from different models relate to the changes of the volumes across different regions in the whole brain. We trained candidate models on the ADNI dataset and tested on three different datasets. Our results indicate: (i) models with higher VCSs tend to demonstrate saliency maps with more details relevant to the AD pathology, (ii) using gradient-based adversarial training strategies such as FGSM and stochastic masking can improve the VCSs of the models.

Haoyang Liu, Yijiang Li, Jinglin Jian et al.

Machine learning has emerged as a powerful tool for scientific discovery, enabling researchers to extract meaningful insights from complex datasets. For instance, it has facilitated the identification of disease-predictive genes from gene expression data, significantly advancing healthcare. However, the traditional process for analyzing such datasets demands substantial human effort and expertise for the data selection, processing, and analysis. To address this challenge, we introduce a novel framework, a Team of AI-made Scientists (TAIS), designed to streamline the scientific discovery pipeline. TAIS comprises simulated roles, including a project manager, data engineer, and domain expert, each represented by a Large Language Model (LLM). These roles collaborate to replicate the tasks typically performed by data scientists, with a specific focus on identifying disease-predictive genes. Furthermore, we have curated a benchmark dataset to assess TAIS's effectiveness in gene identification, demonstrating our system's potential to significantly enhance the efficiency and scope of scientific exploration. Our findings represent a solid step towards automating scientific discovery through large language models.

Peiyan Zhang, Haibo Jin, Leyang Hu et al.

Recent advancements in large language models (LLMs) have significantly enhanced the ability of LLM-based systems to perform complex tasks through natural language processing and tool interaction. However, optimizing these LLM-based systems for specific tasks remains challenging, often requiring manual interventions like prompt engineering and hyperparameter tuning. Existing automatic optimization methods, such as textual feedback-based techniques (e.g., TextGrad), tend to focus on immediate feedback, analogous to using immediate derivatives in traditional numerical gradient descent. However, relying solely on such feedback can be limited when the adjustments made in response to this feedback are either too small or fluctuate irregularly, potentially slowing down or even stalling the optimization process. To overcome these challenges, more adaptive methods are needed, especially in situations where the system's response is evolving slowly or unpredictably. In this paper, we introduce REVOLVE, an optimization method that tracks how "R"esponses "EVOLVE" across iterations in LLM systems. By focusing on the evolution of responses over time, REVOLVE enables more stable and effective optimization by making thoughtful, progressive adjustments at each step. Experimental results demonstrate that REVOLVE outperforms competitive baselines, achieving a 7.8% improvement in prompt optimization, a 20.72% gain in solution refinement, and a 29.17% increase in code optimization. Additionally, REVOLVE converges in fewer iterations, resulting in significant computational savings. Beyond its practical contributions, REVOLVE highlights a promising direction, where the rich knowledge from established optimization principles can be leveraged to enhance LLM systems, which paves the way for further advancements in this hybrid domain.

Jixuan Leng, Chengsong Huang, Langlin Huang et al. · cmu

Existing reasoning evaluation frameworks for Large Language Models (LLMs) and Large Vision-Language Models (LVLMs) predominantly assess either text-based reasoning or vision-language understanding capabilities, with limited dynamic interplay between textual and visual constraints. To address this limitation, we introduce CrossWordBench, a benchmark designed to evaluate the reasoning capabilities of both LLMs and LVLMs through the medium of crossword puzzles -- a task requiring multimodal adherence to semantic constraints from text-based clues and intersectional constraints from visual grid structures. CrossWordBench leverages a controllable puzzle generation framework that produces puzzles in two formats (text and image), supports adjustable difficulty through prefill ratio control, and offers different evaluation strategies, ranging from direct puzzle solving to interactive modes. Our extensive evaluation of over 20 models reveals that reasoning LLMs substantially outperform non-reasoning models by effectively leveraging crossing-letter constraints. We further demonstrate that LVLMs struggle with the task, showing a strong correlation between their puzzle-solving performance and grid-parsing accuracy. Our findings highlight limitations of the reasoning capabilities of current LLMs and LVLMs, and provide an effective approach for creating multimodal constrained tasks for future evaluations.

Yifeng Wang, Ke Chen, Haohan Wang · cmu

Automated diagnosis of Alzheimer Disease(AD) from brain imaging, such as magnetic resonance imaging (MRI), has become increasingly important and has attracted the community to contribute many deep learning methods. However, many of these methods are facing a trade-off that 3D models tend to be complicated while 2D models cannot capture the full 3D intricacies from the data. In this paper, we introduce a new model structure for diagnosing AD, and it can complete with performances of 3D models while essentially is a 2D method (thus computationally efficient). While the core idea lies in new perspective of cutting the 3D images into multiple 2D slices from three dimensions, we introduce multiple components that can further benefit the model in this new perspective, including adaptively selecting the number of sclices in each dimension, and the new attention mechanism. In addition, we also introduce a morphology augmentation, which also barely introduces new computational loads, but can help improve the diagnosis performances due to its alignment to the pathology of AD. We name our method ADAPT, which stands for Alzheimer Diagnosis through Adaptive Profiling Transformers. We test our model from a practical perspective (the testing domains do not appear in the training one): the diagnosis accuracy favors our ADAPT, while ADAPT uses less parameters than most 3D models use.

Yuyang Ji, Haohan Wang

Multimodal Large Language Models (MLLMs) have shown remarkable versatility but face challenges in demonstrating true visual understanding, particularly in chart reasoning tasks. Existing benchmarks like ChartQA reveal significant reliance on text-based shortcuts and probabilistic pattern-matching rather than genuine visual reasoning. To rigorously evaluate visual reasoning, we introduce a more challenging test scenario by removing textual labels and introducing chart perturbations in the ChartQA dataset. Under these conditions, models like GPT-4o and Gemini-2.0 Pro experience up to a 30% performance drop, underscoring their limitations. To address these challenges, we propose Socratic Chart, a new framework that transforms chart images into Scalable Vector Graphics (SVG) representations, enabling MLLMs to integrate textual and visual modalities for enhanced chart understanding. Socratic Chart employs a multi-agent pipeline with specialized agent-generators to extract primitive chart attributes (e.g., bar heights, line coordinates) and an agent-critic to validate results, ensuring high-fidelity symbolic representations. Our framework surpasses state-of-the-art models in accurately capturing chart primitives and improving reasoning performance, establishing a robust pathway for advancing MLLM visual understanding.

Runkai Zheng, Vishnu Asutosh Dasu, Yinong Oliver Wang et al.

Modern machine learning models heavily rely on large datasets that often include sensitive and private information, raising serious privacy concerns. Differentially private (DP) data generation offers a solution by creating synthetic datasets that limit the leakage of private information within a predefined privacy budget; however, it requires a substantial amount of data to achieve performance comparable to models trained on the original data. To mitigate the significant expense incurred with synthetic data generation, Dataset Distillation (DD) stands out for its remarkable training and storage efficiency. This efficiency is particularly advantageous when integrated with DP mechanisms, curating compact yet informative synthetic datasets without compromising privacy. However, current state-of-the-art private DD methods suffer from a synchronized sampling-optimization process and the dependency on noisy training signals from randomly initialized networks. This results in the inefficient utilization of private information due to the addition of excessive noise. To address these issues, we introduce a novel framework that decouples sampling from optimization for better convergence and improves signal quality by mitigating the impact of DP noise through matching in an informative subspace. On CIFAR-10, our method achieves a \textbf{10.0\%} improvement with 50 images per class and \textbf{8.3\%} increase with just \textbf{one-fifth} the distilled set size of previous state-of-the-art methods, demonstrating significant potential to advance privacy-preserving DD.

Peiran Wang, Haohan Wang

In this paper, we introduce DistDD, a novel approach within the federated learning framework that reduces the need for repetitive communication by distilling data directly on clients' devices. Unlike traditional federated learning that requires iterative model updates across nodes, DistDD facilitates a one-time distillation process that extracts a global distilled dataset, maintaining the privacy standards of federated learning while significantly cutting down communication costs. By leveraging the DistDD's distilled dataset, the developers of the FL can achieve just-in-time parameter tuning and neural architecture search over FL without repeating the whole FL process multiple times. We provide a detailed convergence proof of the DistDD algorithm, reinforcing its mathematical stability and reliability for practical applications. Our experiments demonstrate the effectiveness and robustness of DistDD, particularly in non-i.i.d. and mislabeled data scenarios, showcasing its potential to handle complex real-world data challenges distinctively from conventional federated learning methods. We also evaluate DistDD's application in the use case and prove its effectiveness and communication-savings in the NAS use case.

Ke Chen, Yifeng Wang, Hassan Almosapeeh et al.

Most prompt-optimization methods refine a single static template, making them ineffective in complex and dynamic user scenarios. Existing query-dependent approaches rely on unstable textual feedback or black-box reward models, providing weak and uninterpretable optimization signals. More fundamentally, prompt quality itself lacks a unified, systematic definition, resulting in fragmented and unreliable evaluation signals. Our approach first establishes a performance-oriented, systematic, and comprehensive prompt evaluation framework. Furthermore, we develop and finetune an execution-free evaluator that predicts multi-dimensional quality scores directly from text. The evaluator then instructs a metric-aware optimizer that diagnoses failure modes and rewrites prompts in an interpretable, query-dependent manner. Our evaluator achieves the strongest accuracy in predicting prompt performance, and the evaluation-instructed optimization consistently surpass both static-template and query-dependent baselines across eight datasets and on three backbone models. Overall, we propose a unified, metric-grounded perspective on prompt quality, and demonstrated that our evaluation-instructed optimization pipeline delivers stable, interpretable, and model-agnostic improvements across diverse tasks.

Peng Kuang, Yanli Wang, Xiaoyu Han et al.

Process reward models (PRMs) are a cornerstone of test-time scaling (TTS), designed to verify and select the best responses from large language models (LLMs). However, this promise is challenged by recent benchmarks where simple majority voting, which ignores PRM signals, occasionally outperforms standard PRM-based selection. This raises a critical question: How can we effectively utilize verification signals from PRMs for TTS? To address this, we start by developing a theoretical framework for optimally combining signals from both the LLM and the PRM. Our framework reveals that the optimal strategy is a weighted aggregation of responses, a strategy whose effectiveness hinges on estimating weights that capture the complex interplay between the models. Based on our theoretical results, we empirically show that these optimal weighting functions differ significantly across LLM-PRM pairs and, notably, often assign substantial negative weights. Motivated by these insights, we propose efficient pre-computation methods to calibrate these weighting functions. Extensive experiments across 5 LLMs and 7 PRMs demonstrate that our calibration method significantly boosts the TTS efficiency, surpassing the performance of vanilla weighted majority voting while using only $21.3\%$ of the computation. Ultimately, our work demonstrates that investing in a more intelligent aggregation strategy can be a more convincing path to performance gains than simply scaling test-time computation.

Yaoning Yu, Ye Yu, Kai Wei et al.

Prompt quality plays a critical role in the performance of large language models (LLMs), motivating a growing body of work on prompt optimization. Most existing methods optimize prompts over a fixed dataset, assuming static input distributions and offering limited support for iterative improvement. We introduce SIPDO (Self-Improving Prompts through Data-Augmented Optimization), a closed-loop framework for prompt learning that integrates synthetic data generation into the optimization process. SIPDO couples a synthetic data generator with a prompt optimizer, where the generator produces new examples that reveal current prompt weaknesses and the optimizer incrementally refines the prompt in response. This feedback-driven loop enables systematic improvement of prompt performance without assuming access to external supervision or new tasks. Experiments across question answering and reasoning benchmarks show that SIPDO outperforms standard prompt tuning methods, highlighting the value of integrating data synthesis into prompt learning workflows.

Xuezhi Wang, Haohan Wang, Diyi Yang

As NLP models achieved state-of-the-art performances over benchmarks and gained wide applications, it has been increasingly important to ensure the safe deployment of these models in the real world, e.g., making sure the models are robust against unseen or challenging scenarios. Despite robustness being an increasingly studied topic, it has been separately explored in applications like vision and NLP, with various definitions, evaluation and mitigation strategies in multiple lines of research. In this paper, we aim to provide a unifying survey of how to define, measure and improve robustness in NLP. We first connect multiple definitions of robustness, then unify various lines of work on identifying robustness failures and evaluating models' robustness. Correspondingly, we present mitigation strategies that are data-driven, model-driven, and inductive-prior-based, with a more systematic view of how to effectively improve robustness in NLP models. Finally, we conclude by outlining open challenges and future directions to motivate further research in this area.

Haohan Wang, Bryon Aragam, Eric Xing

Motivated by empirical arguments that are well-known from the genome-wide association studies (GWAS) literature, we study the statistical properties of linear mixed models (LMMs) applied to GWAS. First, we study the sensitivity of LMMs to the inclusion of a candidate SNP in the kinship matrix, which is often done in practice to speed up computations. Our results shed light on the size of the error incurred by including a candidate SNP, providing a justification to this technique in order to trade-off velocity against veracity. Second, we investigate how mixed models can correct confounders in GWAS, which is widely accepted as an advantage of LMMs over traditional methods. We consider two sources of confounding factors, population stratification and environmental confounding factors, and study how different methods that are commonly used in practice trade-off these two confounding factors differently.

Xuefeng Du, Haohan Wang, Zhenxi Zhu et al.

Motivation: Cryo-Electron Tomography (cryo-ET) is a 3D bioimaging tool that visualizes the structural and spatial organization of macromolecules at a near-native state in single cells, which has broad applications in life science. However, the systematic structural recognition and recovery of macromolecules captured by cryo-ET are difficult due to high structural complexity and imaging limits. Deep learning based subtomogram classification have played critical roles for such tasks. As supervised approaches, however, their performance relies on sufficient and laborious annotation on a large training dataset. Results: To alleviate this major labeling burden, we proposed a Hybrid Active Learning (HAL) framework for querying subtomograms for labelling from a large unlabeled subtomogram pool. Firstly, HAL adopts uncertainty sampling to select the subtomograms that have the most uncertain predictions. Moreover, to mitigate the sampling bias caused by such strategy, a discriminator is introduced to judge if a certain subtomogram is labeled or unlabeled and subsequently the model queries the subtomogram that have higher probabilities to be unlabeled. Additionally, HAL introduces a subset sampling strategy to improve the diversity of the query set, so that the information overlap is decreased between the queried batches and the algorithmic efficiency is improved. Our experiments on subtomogram classification tasks using both simulated and real data demonstrate that we can achieve comparable testing performance (on average only 3% accuracy drop) by using less than 30% of the labeled subtomograms, which shows a very promising result for subtomogram classification task with limited labeling resources.

He He, Sheng Zha, Haohan Wang

Statistical natural language inference (NLI) models are susceptible to learning dataset bias: superficial cues that happen to associate with the label on a particular dataset, but are not useful in general, e.g., negation words indicate contradiction. As exposed by several recent challenge datasets, these models perform poorly when such association is absent, e.g., predicting that "I love dogs" contradicts "I don't love cats". Our goal is to design learning algorithms that guard against known dataset bias. We formalize the concept of dataset bias under the framework of distribution shift and present a simple debiasing algorithm based on residual fitting, which we call DRiFt. We first learn a biased model that only uses features that are known to relate to dataset bias. Then, we train a debiased model that fits to the residual of the biased model, focusing on examples that cannot be predicted well by biased features only. We use DRiFt to train three high-performing NLI models on two benchmark datasets, SNLI and MNLI. Our debiased models achieve significant gains over baseline models on two challenge test sets, while maintaining reasonable performance on the original test sets.

Haohan Wang, Xindi Wu, Zeyi Huang et al.

We investigate the relationship between the frequency spectrum of image data and the generalization behavior of convolutional neural networks (CNN). We first notice CNN's ability in capturing the high-frequency components of images. These high-frequency components are almost imperceptible to a human. Thus the observation leads to multiple hypotheses that are related to the generalization behaviors of CNN, including a potential explanation for adversarial examples, a discussion of CNN's trade-off between robustness and accuracy, and some evidence in understanding training heuristics.

Haohan Wang, Da Sun, Eric P. Xing

Nature language inference (NLI) task is a predictive task of determining the inference relationship of a pair of natural language sentences. With the increasing popularity of NLI, many state-of-the-art predictive models have been proposed with impressive performances. However, several works have noticed the statistical irregularities in the collected NLI data set that may result in an over-estimated performance of these models and proposed remedies. In this paper, we further investigate the statistical irregularities, what we refer as confounding factors, of the NLI data sets. With the belief that some NLI labels should preserve under swapping operations, we propose a simple yet effective way (swapping the two text fragments) of evaluating the NLI predictive models that naturally mitigate the observed problems. Further, we continue to train the predictive models with our swapping manner and propose to use the deviation of the model's evaluation performances under different percentages of training text fragments to be swapped to describe the robustness of a predictive model. Our evaluation metrics leads to some interesting understandings of recent published NLI methods. Finally, we also apply the swapping operation on NLI models to see the effectiveness of this straightforward method in mitigating the confounding factor problems in training generic sentence embeddings for other NLP transfer tasks.

Haohan Wang, Zhenglin Wu, Eric P. Xing

The proliferation of healthcare data has brought the opportunities of applying data-driven approaches, such as machine learning methods, to assist diagnosis. Recently, many deep learning methods have been shown with impressive successes in predicting disease status with raw input data. However, the "black-box" nature of deep learning and the high-reliability requirement of biomedical applications have created new challenges regarding the existence of confounding factors. In this paper, with a brief argument that inappropriate handling of confounding factors will lead to models' sub-optimal performance in real-world applications, we present an efficient method that can remove the influences of confounding factors such as age or gender to improve the across-cohort prediction accuracy of neural networks. One distinct advantage of our method is that it only requires minimal changes of the baseline model's architecture so that it can be plugged into most of the existing neural networks. We conduct experiments across CT-scan, MRA, and EEG brain wave with convolutional neural networks and LSTM to verify the efficiency of our method.

Haohan Wang, Bhiksha Raj

This paper is a review of the evolutionary history of deep learning models. It covers from the genesis of neural networks when associationism modeling of the brain is studied, to the models that dominate the last decade of research in deep learning like convolutional neural networks, deep belief networks, and recurrent neural networks. In addition to a review of these models, this paper primarily focuses on the precedents of the models above, examining how the initial ideas are assembled to construct the early models and how these preliminary models are developed into their current forms. Many of these evolutionary paths last more than half a century and have a diversity of directions. For example, CNN is built on prior knowledge of biological vision system; DBN is evolved from a trade-off of modeling power and computation complexity of graphical models and many nowadays models are neural counterparts of ancient linear models. This paper reviews these evolutionary paths and offers a concise thought flow of how these models are developed, and aims to provide a thorough background for deep learning. More importantly, along with the path, this paper summarizes the gist behind these milestones and proposes many directions to guide the future research of deep learning.

Haohan Wang, Aaksha Meghawat, Louis-Philippe Morency et al.

Multimodal sentiment analysis is drawing an increasing amount of attention these days. It enables mining of opinions in video reviews which are now available aplenty on online platforms. However, multimodal sentiment analysis has only a few high-quality data sets annotated for training machine learning algorithms. These limited resources restrict the generalizability of models, where, for example, the unique characteristics of a few speakers (e.g., wearing glasses) may become a confounding factor for the sentiment classification task. In this paper, we propose a Select-Additive Learning (SAL) procedure that improves the generalizability of trained neural networks for multimodal sentiment analysis. In our experiments, we show that our SAL approach improves prediction accuracy significantly in all three modalities (verbal, acoustic, visual), as well as in their fusion. Our results show that SAL, even when trained on one dataset, achieves good generalization across two new test datasets.

Haohan Wang, Madhavi K. Ganapathiraju

In order for the predicted interactions to be directly adopted by biologists, the ma- chine learning predictions have to be of high precision, regardless of recall. This aspect cannot be evaluated or numerically represented well by traditional metrics like accuracy, ROC, or precision-recall curve. In this work, we start from the alignment in sensitivity of ROC and recall of precision-recall curve, and propose an evaluation metric focusing on the ability of a model to be adopted by biologists. This metric evaluates the ability of a machine learning algorithm to predict only new interactions, meanwhile, it eliminates the influence of test dataset. In the experiment of evaluating different classifiers with a same data set and evaluating the same predictor with different datasets, our new metric fulfills the evaluation task of our interest while two widely recognized metrics, ROC and precision-recall curve fail the tasks for different reasons.

Haohan Wang, Bhiksha Raj

This report will show the history of deep learning evolves. It will trace back as far as the initial belief of connectionism modelling of brain, and come back to look at its early stage realization: neural networks. With the background of neural network, we will gradually introduce how convolutional neural network, as a representative of deep discriminative models, is developed from neural networks, together with many practical techniques that can help in optimization of neural networks. On the other hand, we will also trace back to see the evolution history of deep generative models, to see how researchers balance the representation power and computation complexity to reach Restricted Boltzmann Machine and eventually reach Deep Belief Nets. Further, we will also look into the development history of modelling time series data with neural networks. We start with Time Delay Neural Networks and move further to currently famous model named Recurrent Neural Network and its extension Long Short Term Memory. We will also briefly look into how to construct deep recurrent neural networks. Finally, we will conclude this report with some interesting open-ended questions of deep neural networks.

Haohan Wang, Madhavi K. Ganapathiraju

Protein-protein interaction (PPI) prediction is an important problem in machine learning and computational biology. However, there is no data set for training or evaluation purposes, where all the instances are accurately labeled. Instead, what is available are instances of positive class (with possibly noisy labels) and no instances of negative class. The non-availability of negative class data is typically handled with the observation that randomly chosen protein-pairs have a nearly 100% chance of being negative class, as only 1 in 1,500 protein pairs expected is expected to be an interacting pair. In this paper, we focused on the problem that non-availability of accurately labeled testing data sets in the domain of protein-protein interaction (PPI) prediction may lead to biased evaluation results. We first showed that not acknowledging the inherent skew in the interactome (i.e. rare occurrence of positive instances) leads to an over-estimated accuracy of the predictor. Then we show that, with the belief that positive interactions are a rare category, sampling random pairs of proteins excluding known interacting proteins set as the negative testing data set could lead to an under-estimated evaluation result. We formalized those two problems to validate the above claim, and based on the formalization, we proposed a balancing method to cancel out the over-estimation with under-estimation. Finally, our experiments validated the theoretical aspects and showed that this balancing evaluation could evaluate the exact performance without availability of golden standard data sets.

Seungwhan Moon, Suyoun Kim, Haohan Wang

We propose a transfer deep learning (TDL) framework that can transfer the knowledge obtained from a single-modal neural network to a network with a different modality. Specifically, we show that we can leverage speech data to fine-tune the network trained for video recognition, given an initial set of audio-video parallel dataset within the same semantics. Our approach first learns the analogy-preserving embeddings between the abstract representations learned from intermediate layers of each network, allowing for semantics-level transfer between the source and target modalities. We then apply our neural network operation that fine-tunes the target network with the additional knowledge transferred from the source network, while keeping the topology of the target network unchanged. While we present an audio-visual recognition task as an application of our approach, our framework is flexible and thus can work with any multimodal dataset, or with any already-existing deep networks that share the common underlying semantics. In this work in progress report, we aim to provide comprehensive results of different configurations of the proposed approach on two widely used audio-visual datasets, and we discuss potential applications of the proposed approach.

Ming Xu, Jianping Wu, Yiman Du et al.

A major problem in road network analysis is discovery of important crossroads, which can provide useful information for transport planning. However, none of existing approaches addresses the problem of identifying network-wide important crossroads in real road network. In this paper, we propose a novel data-driven based approach named CRRank to rank important crossroads. Our key innovation is that we model the trip network reflecting real travel demands with a tripartite graph, instead of solely analysis on the topology of road network. To compute the importance scores of crossroads accurately, we propose a HITS-like ranking algorithm, in which a procedure of score propagation on our tripartite graph is performed. We conduct experiments on CRRank using a real-world dataset of taxi trajectories. Experiments verify the utility of CRRank.