Kaiwen Zha

Kaiwen Zha, Peng Cao, Jeany Son et al.

Deep regression models typically learn in an end-to-end fashion without explicitly emphasizing a regression-aware representation. Consequently, the learned representations exhibit fragmentation and fail to capture the continuous nature of sample orders, inducing suboptimal results across a wide range of regression tasks. To fill the gap, we propose Rank-N-Contrast (RNC), a framework that learns continuous representations for regression by contrasting samples against each other based on their rankings in the target space. We demonstrate, theoretically and empirically, that RNC guarantees the desired order of learned representations in accordance with the target orders, enjoying not only better performance but also significantly improved robustness, efficiency, and generalization. Extensive experiments using five real-world regression datasets that span computer vision, human-computer interaction, and healthcare verify that RNC achieves state-of-the-art performance, highlighting its intriguing properties including better data efficiency, robustness to spurious targets and data corruptions, and generalization to distribution shifts. Code is available at: https://github.com/kaiwenzha/Rank-N-Contrast.

Kaiwen Zha, Zhengqi Gao, Maohao Shen et al.

Reinforcement learning (RL) has recently emerged as a compelling approach for enhancing the reasoning capabilities of large language models (LLMs), where an LLM generator serves as a policy guided by a verifier (reward model). However, current RL post-training methods for LLMs typically use verifiers that are fixed (rule-based or frozen pretrained) or trained discriminatively via supervised fine-tuning (SFT). Such designs are susceptible to reward hacking and generalize poorly beyond their training distributions. To overcome these limitations, we propose Tango, a novel framework that uses RL to concurrently train both an LLM generator and a verifier in an interleaved manner. A central innovation of Tango is its generative, process-level LLM verifier, which is trained via RL and co-evolves with the generator. Importantly, the verifier is trained solely based on outcome-level verification correctness rewards without requiring explicit process-level annotations. This generative RL-trained verifier exhibits improved robustness and superior generalization compared to deterministic or SFT-trained verifiers, fostering effective mutual reinforcement with the generator. Extensive experiments demonstrate that both components of Tango achieve state-of-the-art results among 7B/8B-scale models: the generator attains best-in-class performance across five competition-level math benchmarks and four challenging out-of-domain reasoning tasks, while the verifier leads on the ProcessBench dataset. Remarkably, both components exhibit particularly substantial improvements on the most difficult mathematical reasoning problems. Code is at: https://github.com/kaiwenzha/rl-tango.

Siru Ouyang, Jun Yan, Yanfei Chen et al.

LLM-based agents are increasingly deployed to handle streaming tasks, yet they often remain one-off problem solvers that fail to learn from past interactions. Reusable skills distilled from experience provide a natural substrate for self-evolution, where high-quality skill curation serves as the key bottleneck. Existing approaches either rely on manual skill curation, prescribe heuristic skill operations, or train for short-horizon skill operations. However, they still struggle to learn complex long-term curation policies from indirect and delayed feedback. To tackle this challenge, we propose SkillOS, an experience-driven RL training recipe for learning skill curation in self-evolving agents. SkillOS pairs a frozen agent executor that retrieves and applies skills with a trainable skill curator that updates an external SkillRepo from accumulated experience. To provide learning signals for curation, we design composite rewards and train on grouped task streams based on skill-relevant task dependencies, where earlier trajectories update the SkillRepo, and later related tasks evaluate these updates. Across multi-turn agentic tasks and single-turn reasoning tasks, SkillOS consistently outperforms memory-free and strong memory-based baselines in both effectiveness and efficiency, with the learned skill curator generalizing across different executor backbones and task domains. Further analyses show that the learned curator produces more targeted skill use, while the skills in SkillRepo evolve into more richly structured Markdown files that encode higher-level meta-skills over time.

Hao He, Kaiwen Zha, Dina Katabi

Indiscriminate data poisoning attacks are quite effective against supervised learning. However, not much is known about their impact on unsupervised contrastive learning (CL). This paper is the first to consider indiscriminate poisoning attacks of contrastive learning. We propose Contrastive Poisoning (CP), the first effective such attack on CL. We empirically show that Contrastive Poisoning, not only drastically reduces the performance of CL algorithms, but also attacks supervised learning models, making it the most generalizable indiscriminate poisoning attack. We also show that CL algorithms with a momentum encoder are more robust to indiscriminate poisoning, and propose a new countermeasure based on matrix completion. Code is available at: https://github.com/kaiwenzha/contrastive-poisoning.

Yuzhe Yang, Kaiwen Zha, Ying-Cong Chen et al.

Real-world data often exhibit imbalanced distributions, where certain target values have significantly fewer observations. Existing techniques for dealing with imbalanced data focus on targets with categorical indices, i.e., different classes. However, many tasks involve continuous targets, where hard boundaries between classes do not exist. We define Deep Imbalanced Regression (DIR) as learning from such imbalanced data with continuous targets, dealing with potential missing data for certain target values, and generalizing to the entire target range. Motivated by the intrinsic difference between categorical and continuous label space, we propose distribution smoothing for both labels and features, which explicitly acknowledges the effects of nearby targets, and calibrates both label and learned feature distributions. We curate and benchmark large-scale DIR datasets from common real-world tasks in computer vision, natural language processing, and healthcare domains. Extensive experiments verify the superior performance of our strategies. Our work fills the gap in benchmarks and techniques for practical imbalanced regression problems. Code and data are available at https://github.com/YyzHarry/imbalanced-regression.

Bo Pang, Kaiwen Zha, Hanwen Cao et al.

Extracting temporal and representation features efficiently plays a pivotal role in understanding visual sequence information. To deal with this, we propose a new recurrent neural framework that can be stacked deep effectively. There are mainly two novel designs in our deep RNN framework: one is a new RNN module called Context Bridge Module (CBM) which splits the information flowing along the sequence (temporal direction) and along depth (spatial representation direction), making it easier to train when building deep by balancing these two directions; the other is the Overlap Coherence Training Scheme that reduces the training complexity for long visual sequential tasks on account of the limitation of computing resources. We provide empirical evidence to show that our deep RNN framework is easy to optimize and can gain accuracy from the increased depth on several visual sequence problems. On these tasks, we evaluate our deep RNN framework with 15 layers, 7* than conventional RNN networks, but it is still easy to train. Our deep framework achieves more than 11% relative improvements over shallow RNN models on Kinetics, UCF-101, and HMDB-51 for video classification. For auxiliary annotation, after replacing the shallow RNN part of Polygon-RNN with our 15-layer deep CBM, the performance improves by 14.7%. For video future prediction, our deep RNN improves the state-of-the-art shallow model's performance by 2.4% on PSNR and SSIM. The code and trained models are published accompanied by this paper: https://github.com/BoPang1996/Deep-RNN-Framework.

Kaiwen Zha, Lijun Yu, Alireza Fathi et al.

Image tokenization, the process of transforming raw image pixels into a compact low-dimensional latent representation, has proven crucial for scalable and efficient image generation. However, mainstream image tokenization methods generally have limited compression rates, making high-resolution image generation computationally expensive. To address this challenge, we propose to leverage language for efficient image tokenization, and we call our method Text-Conditioned Image Tokenization (TexTok). TexTok is a simple yet effective tokenization framework that leverages language to provide a compact, high-level semantic representation. By conditioning the tokenization process on descriptive text captions, TexTok simplifies semantic learning, allowing more learning capacity and token space to be allocated to capture fine-grained visual details, leading to enhanced reconstruction quality and higher compression rates. Compared to the conventional tokenizer without text conditioning, TexTok achieves average reconstruction FID improvements of 29.2% and 48.1% on ImageNet-256 and -512 benchmarks respectively, across varying numbers of tokens. These tokenization improvements consistently translate to 16.3% and 34.3% average improvements in generation FID. By simply replacing the tokenizer in Diffusion Transformer (DiT) with TexTok, our system can achieve a 93.5x inference speedup while still outperforming the original DiT using only 32 tokens on ImageNet-512. TexTok with a vanilla DiT generator achieves state-of-the-art FID scores of 1.46 and 1.62 on ImageNet-256 and -512 respectively. Furthermore, we demonstrate TexTok's superiority on the text-to-image generation task, effectively utilizing the off-the-shelf text captions in tokenization. Project page is at: https://kaiwenzha.github.io/textok/.

Zhengqi Gao, Kaiwen Zha, Tianyuan Zhang et al.

Guidance techniques are simple yet effective for improving conditional generation in diffusion models. Albeit their empirical success, the practical implementation of guidance diverges significantly from its theoretical motivation. In this paper, we reconcile this discrepancy by replacing the scaled marginal distribution target, which we prove theoretically invalid, with a valid scaled joint distribution objective. Additionally, we show that the established guidance implementations are approximations to the intractable optimal solution under no future foresight constraint. Building on these theoretical insights, we propose rectified gradient guidance (REG), a versatile enhancement designed to boost the performance of existing guidance methods. Experiments on 1D and 2D demonstrate that REG provides a better approximation to the optimal solution than prior guidance techniques, validating the proposed theoretical framework. Extensive experiments on class-conditional ImageNet and text-to-image generation tasks show that incorporating REG consistently improves FID and Inception/CLIP scores across various settings compared to its absence.

Maohao Shen, Kaiwen Zha, Zexue He et al.

There is growing interest in improving LLMs without updating model parameters. One well-established direction is test-time scaling, where increased inference-time computation (e.g., longer reasoning, sampling, or search) is used to improve performance. However, for complex reasoning and agentic tasks, naively scaling test-time compute can substantially increase cost and still lead to wasted budget on suboptimal exploration. In this paper, we explore \emph{context} as a complementary scaling axis for improving LLM performance, and systematically study how to construct better inputs that guide reasoning through \emph{experience}. We show that effective context construction critically depends on \emph{decocted experience}. We present a detailed analysis of experience-augmented agents, studying how to derive context from experience, how performance scales with accumulated experience, what characterizes good context, and which data structures best support context construction. We identify \emph{decocted experience} as a key mechanism for effective context construction: extracting essence from experience, organizing it coherently, and retrieving salient information to build effective context. We validate our findings across reasoning and agentic tasks, including math reasoning, web browsing, and software engineering.

Ali Mirzazadeh, Simon Cadavid, Kaiwen Zha et al.

Antidepressant nonadherence is pervasive, driving relapse, hospitalization, suicide risk, and billions in avoidable costs. Clinicians need tools that detect adherence lapses promptly, yet current methods are either invasive (serum assays, neuroimaging) or proxy-based and inaccurate (pill counts, pharmacy refills). We present the first noninvasive biomarker that detects antidepressant intake from a single night of sleep. A transformer-based model analyzes sleep data from a consumer wearable or contactless wireless sensor to infer antidepressant intake, enabling remote, effortless, daily adherence assessment at home. Across six datasets comprising 62,000 nights from >20,000 participants (1,800 antidepressant users), the biomarker achieved AUROC = 0.84, generalized across drug classes, scaled with dose, and remained robust to concomitant psychotropics. Longitudinal monitoring captured real-world initiation, tapering, and lapses. This approach offers objective, scalable adherence surveillance with potential to improve depression care and outcomes.

Kaiwen Zha, Yujun Shen, Bolei Zhou

In this work, we study the image transformation problem, which targets at learning the underlying transformations (e.g., the transition of seasons) from a collection of unlabeled images. However, there could be countless of transformations in the real world, making such a task incredibly challenging, especially under the unsupervised setting. To tackle this obstacle, we propose a novel learning framework built on generative adversarial networks (GANs), where the discriminator and the generator share a transformation space. After the model gets fully optimized, any two points within the shared space are expected to define a valid transformation. In this way, at the inference stage, we manage to adequately extract the variation factor between a customizable image pair by projecting both images onto the transformation space. The resulting transformation vector can further guide the image synthesis, facilitating image editing with continuous semantic change (e.g., altering summer to winter with fall as the intermediate step). Noticeably, the learned transformation space supports not only transferring image styles (e.g., changing day to night), but also manipulating image contents (e.g., adding clouds in the sky). In addition, we make in-depth analysis on the properties of the transformation space to help understand how various transformations are organized. Project page is at https://genforce.github.io/trgan/.

Bo Pang, Kaiwen Zha, Hanwen Cao et al.

Understanding sequential information is a fundamental task for artificial intelligence. Current neural networks attempt to learn spatial and temporal information as a whole, limited their abilities to represent large scale spatial representations over long-range sequences. Here, we introduce a new modeling strategy called Semi-Coupled Structure (SCS), which consists of deep neural networks that decouple the complex spatial and temporal concepts learning. Semi-Coupled Structure can learn to implicitly separate input information into independent parts and process these parts respectively. Experiments demonstrate that a Semi-Coupled Structure can successfully annotate the outline of an object in images sequentially and perform video action recognition. For sequence-to-sequence problems, a Semi-Coupled Structure can predict future meteorological radar echo images based on observed images. Taken together, our results demonstrate that a Semi-Coupled Structure has the capacity to improve the performance of LSTM-like models on large scale sequential tasks.

Bo Pang, Kaiwen Zha, Cewu Lu

We introduce the first benchmark for a new problem --- recognizing human action adverbs (HAA): "Adverbs Describing Human Actions" (ADHA). This is the first step for computer vision to change over from pattern recognition to real AI. We demonstrate some key features of ADHA: a semantically complete set of adverbs describing human actions, a set of common, describable human actions, and an exhaustive labeling of simultaneously emerging actions in each video. We commit an in-depth analysis on the implementation of current effective models in action recognition and image captioning on adverb recognition, and the results show that such methods are unsatisfactory. Moreover, we propose a novel three-stream hybrid model to deal the HAA problem, which achieves a better result.