Zhihan Yang

Yuanyuan Wang, Yang Zhang, Zhiyong Wu et al.

Data augmentation is vital to the generalization ability and robustness of deep neural networks (DNNs) models. Existing augmentation methods for speaker verification manipulate the raw signal, which are time-consuming and the augmented samples lack diversity. In this paper, we present a novel difficulty-aware semantic augmentation (DASA) approach for speaker verification, which can generate diversified training samples in speaker embedding space with negligible extra computing cost. Firstly, we augment training samples by perturbing speaker embeddings along semantic directions, which are obtained from speaker-wise covariance matrices. Secondly, accurate covariance matrices are estimated from robust speaker embeddings during training, so we introduce difficultyaware additive margin softmax (DAAM-Softmax) to obtain optimal speaker embeddings. Finally, we assume the number of augmented samples goes to infinity and derive a closed-form upper bound of the expected loss with DASA, which achieves compatibility and efficiency. Extensive experiments demonstrate the proposed approach can achieve a remarkable performance improvement. The best result achieves a 14.6% relative reduction in EER metric on CN-Celeb evaluation set.

Zhihan Yang, Zhiming Cheng, Tengjin Weng et al.

Retinal disease is one of the primary causes of visual impairment, and early diagnosis is essential for preventing further deterioration. Nowadays, many works have explored Transformers for diagnosing diseases due to their strong visual representation capabilities. However, retinal diseases exhibit milder forms and often present with overlapping signs, which pose great difficulties for accurate multi-class classification. Therefore, we propose a new framework named Multi-Scale Patch Message Passing Swin Transformer for multi-class retinal disease classification. Specifically, we design a Patch Message Passing (PMP) module based on the Message Passing mechanism to establish global interaction for pathological semantic features and to exploit the subtle differences further between different diseases. Moreover, considering the various scale of pathological features we integrate multiple PMP modules for different patch sizes. For evaluation, we have constructed a new dataset, named OPTOS dataset, consisting of 1,033 high-resolution fundus images photographed by Optos camera and conducted comprehensive experiments to validate the efficacy of our proposed method. And the results on both the public dataset and our dataset demonstrate that our method achieves remarkable performance compared to state-of-the-art methods.

Zhihan Yang, Hai Nguyen

When the environment is partially observable (PO), a deep reinforcement learning (RL) agent must learn a suitable temporal representation of the entire history in addition to a strategy to control. This problem is not novel, and there have been model-free and model-based algorithms proposed for this problem. However, inspired by recent success in model-free image-based RL, we noticed the absence of a model-free baseline for history-based RL that (1) uses full history and (2) incorporates recent advances in off-policy continuous control. Therefore, we implement recurrent versions of DDPG, TD3, and SAC (RDPG, RTD3, and RSAC) in this work, evaluate them on short-term and long-term PO domains, and investigate key design choices. Our experiments show that RDPG and RTD3 can surprisingly fail on some domains and that RSAC is the most reliable, reaching near-optimal performance on nearly all domains. However, one task that requires systematic exploration still proved to be difficult, even for RSAC. These results show that model-free RL can learn good temporal representation using only reward signals; the primary difficulty seems to be computational cost and exploration. To facilitate future research, we have made our PyTorch implementation publicly available at https://github.com/zhihanyang2022/off-policy-continuous-control.

Guangqian Yang, Kangrui Du, Zhihan Yang et al.

Alzheimer's disease (AD) is an incurable neurodegenerative condition leading to cognitive and functional deterioration. Given the lack of a cure, prompt and precise AD diagnosis is vital, a complex process dependent on multiple factors and multi-modal data. While successful efforts have been made to integrate multi-modal representation learning into medical datasets, scant attention has been given to 3D medical images. In this paper, we propose Contrastive Masked Vim Autoencoder (CMViM), the first efficient representation learning method tailored for 3D multi-modal data. Our proposed framework is built on a masked Vim autoencoder to learn a unified multi-modal representation and long-dependencies contained in 3D medical images. We also introduce an intra-modal contrastive learning module to enhance the capability of the multi-modal Vim encoder for modeling the discriminative features in the same modality, and an inter-modal contrastive learning module to alleviate misaligned representation among modalities. Our framework consists of two main steps: 1) incorporate the Vision Mamba (Vim) into the mask autoencoder to reconstruct 3D masked multi-modal data efficiently. 2) align the multi-modal representations with contrastive learning mechanisms from both intra-modal and inter-modal aspects. Our framework is pre-trained and validated ADNI2 dataset and validated on the downstream task for AD classification. The proposed CMViM yields 2.7\% AUC performance improvement compared with other state-of-the-art methods.

Lang Mei, Siyu Mo, Zhihan Yang et al.

Multimodal Retrieval-Augmented Generation (MRAG) enhances large language models (LLMs) by integrating multimodal data (text, images, videos) into retrieval and generation processes, overcoming the limitations of text-only Retrieval-Augmented Generation (RAG). While RAG improves response accuracy by incorporating external textual knowledge, MRAG extends this framework to include multimodal retrieval and generation, leveraging contextual information from diverse data types. This approach reduces hallucinations and enhances question-answering systems by grounding responses in factual, multimodal knowledge. Recent studies show MRAG outperforms traditional RAG, especially in scenarios requiring both visual and textual understanding. This survey reviews MRAG's essential components, datasets, evaluation methods, and limitations, providing insights into its construction and improvement. It also identifies challenges and future research directions, highlighting MRAG's potential to revolutionize multimodal information retrieval and generation. By offering a comprehensive perspective, this work encourages further exploration into this promising paradigm.

Subham Sekhar Sahoo, Zhihan Yang, Yash Akhauri et al.

Diffusion-based language models offer a compelling alternative to autoregressive (AR) models by enabling parallel and controllable generation. Among this family of models, Masked Diffusion Models (MDMs) achieve the strongest performance but still underperform AR models in perplexity and lack key inference-time efficiency features--most notably, KV caching. In this work, we introduce Eso-LMs, a new family of models that fuses AR and MDM paradigms, enabling smooth interpolation between their perplexities while overcoming their respective limitations. Eso-LMs set a new state of the art on standard language modeling benchmarks. Crucially, we are the **first to introduce KV caching for MDMs** while preserving parallel generation, significantly improving inference efficiency. Combined with an optimized sampling schedule, our method achieves up to **65x** faster inference than standard MDMs and **4x** faster inference than prior semi-autoregressive approaches. We provide the code and model checkpoints on the project page: [http://s-sahoo.github.io/Eso-LMs](http://s-sahoo.github.io/Eso-LMs)

Lang Mei, Zhihan Yang, Chong Chen

Recent studies have explored integrating Large Language Models (LLMs) with search engines to leverage both the LLMs' internal pre-trained knowledge and external information. Specially, reinforcement learning (RL) has emerged as a promising paradigm for enhancing LLM reasoning through multi-turn interactions with search engines. However, existing RL-based search agents rely on a single LLM to handle both search planning and question-answering (QA) tasks in an end-to-end manner, which limits their ability to optimize both capabilities simultaneously. In practice, sophisticated AI search systems often employ a large, frozen LLM (e.g., GPT-4, DeepSeek-R1) to ensure high-quality QA. Thus, a more effective and efficient approach is to utilize a small, trainable LLM dedicated to search planning. In this paper, we propose \textbf{AI-SearchPlanner}, a novel reinforcement learning framework designed to enhance the performance of frozen QA models by focusing on search planning. Specifically, our approach introduces three key innovations: 1) Decoupling the Architecture of the Search Planner and Generator, 2) Dual-Reward Alignment for Search Planning, and 3) Pareto Optimization of Planning Utility and Cost, to achieve the objectives. Extensive experiments on real-world datasets demonstrate that AI SearchPlanner outperforms existing RL-based search agents in both effectiveness and efficiency, while exhibiting strong generalization capabilities across diverse frozen QA models and data domains.

Yaohua Bu, Tianyi Ma, Weijun Li et al.

Second language (L2) English learners often find it difficult to improve their pronunciations due to the lack of expressive and personalized corrective feedback. In this paper, we present Pronunciation Teacher (PTeacher), a Computer-Aided Pronunciation Training (CAPT) system that provides personalized exaggerated audio-visual corrective feedback for mispronunciations. Though the effectiveness of exaggerated feedback has been demonstrated, it is still unclear how to define the appropriate degrees of exaggeration when interacting with individual learners. To fill in this gap, we interview 100 L2 English learners and 22 professional native teachers to understand their needs and experiences. Three critical metrics are proposed for both learners and teachers to identify the best exaggeration levels in both audio and visual modalities. Additionally, we incorporate the personalized dynamic feedback mechanism given the English proficiency of learners. Based on the obtained insights, a comprehensive interactive pronunciation training course is designed to help L2 learners rectify mispronunciations in a more perceptible, understandable, and discriminative manner. Extensive user studies demonstrate that our system significantly promotes the learners' learning efficiency.

Anurag Sarkar, Zhihan Yang, Seth Cooper

Prior research has shown variational autoencoders (VAEs) to be useful for generating and blending game levels by learning latent representations of existing level data. We build on such models by exploring the level design affordances and applications enabled by conditional VAEs (CVAEs). CVAEs augment VAEs by allowing them to be trained using labeled data, thus enabling outputs to be generated conditioned on some input. We studied how increased control in the level generation process and the ability to produce desired outputs via training on labeled game level data could build on prior PCGML methods. Through our results of training CVAEs on levels from Super Mario Bros., Kid Icarus and Mega Man, we show that such models can assist in level design by generating levels with desired level elements and patterns as well as producing blended levels with desired combinations of games.

Zhihan Yang, Anurag Sarkar, Seth Cooper

Variational autoencoders (VAEs) have been shown to be able to generate game levels but require manual exploration of the learned latent space to generate outputs with desired attributes. While conditional VAEs address this by allowing generation to be conditioned on labels, such labels have to be provided during training and thus require prior knowledge which may not always be available. In this paper, we apply Gaussian Mixture VAEs (GMVAEs), a variant of the VAE which imposes a mixture of Gaussians (GM) on the latent space, unlike regular VAEs which impose a unimodal Gaussian. This allows GMVAEs to cluster levels in an unsupervised manner using the components of the GM and then generate new levels using the learned components. We demonstrate our approach with levels from Super Mario Bros., Kid Icarus and Mega Man. Our results show that the learned components discover and cluster level structures and patterns and can be used to generate levels with desired characteristics.

Anurag Sarkar, Zhihan Yang, Seth Cooper

Previous work explored blending levels from existing games to create levels for a new game that mixes properties of the original games. In this paper, we use Variational Autoencoders (VAEs) for improving upon such techniques. VAEs are artificial neural networks that learn and use latent representations of datasets to generate novel outputs. We train a VAE on level data from Super Mario Bros. and Kid Icarus, enabling it to capture the latent space spanning both games. We then use this space to generate level segments that combine properties of levels from both games. Moreover, by applying evolutionary search in the latent space, we evolve level segments satisfying specific constraints. We argue that these affordances make the VAE-based approach especially suitable for co-creative level design and compare its performance with similar generative models like the GAN and the VAE-GAN.