Guoshuai Zhao

Houwen Peng, Kan Wu, Yixuan Wei et al.

In this paper, we explore FP8 low-bit data formats for efficient training of large language models (LLMs). Our key insight is that most variables, such as gradients and optimizer states, in LLM training can employ low-precision data formats without compromising model accuracy and requiring no changes to hyper-parameters. Specifically, we propose a new FP8 automatic mixed-precision framework for training LLMs. This framework offers three levels of FP8 utilization to streamline mixed-precision and distributed parallel training for LLMs. It gradually incorporates 8-bit gradients, optimizer states, and distributed learning in an incremental manner. Experiment results show that, during the training of GPT-175B model on H100 GPU platform, our FP8 mixed-precision training framework not only achieved a remarkable 39% reduction in real memory usage but also ran 75% faster than the widely adopted BF16 framework (i.e., Megatron-LM), surpassing the speed of Nvidia Transformer Engine by 37%. This largely reduces the training costs for large foundation models. Furthermore, our FP8 mixed-precision training methodology is generic. It can be seamlessly applied to other tasks such as LLM instruction tuning and reinforcement learning with human feedback, offering savings in fine-tuning expenses. Our FP8 low-precision training framework is open-sourced at {https://github.com/Azure/MS-AMP}{aka.ms/MS.AMP}.

Ye Wang, Yaxiong Wang, Guoshuai Zhao et al.

Few-shot class-incremental learning (FSCIL) aims to continually learn new classes using a few samples while not forgetting the old classes. The key of this task is effective knowledge transfer from the base session to the incremental sessions. Despite the advance of existing FSCIL methods, the proposed knowledge transfer learning schemes are sub-optimal due to the insufficient optimization for the model's plasticity. To address this issue, we propose a Random Episode Sampling and Augmentation (RESA) strategy that relies on diverse pseudo incremental tasks as agents to achieve the knowledge transfer. Concretely, RESA mimics the real incremental setting and constructs pseudo incremental tasks globally and locally, where the global pseudo incremental tasks are designed to coincide with the learning objective of FSCIL and the local pseudo incremental tasks are designed to improve the model's plasticity, respectively. Furthermore, to make convincing incremental predictions, we introduce a complementary model with a squared Euclidean-distance classifier as the auxiliary module, which couples with the widely used cosine classifier to form our whole architecture. By such a way, equipped with model decoupling strategy, we can maintain the model's stability while enhancing the model's plasticity. Extensive quantitative and qualitative experiments on three popular FSCIL benchmark datasets demonstrate that our proposed method, named Knowledge Transfer-driven Relation Complementation Network (KT-RCNet), outperforms almost all prior methods. More precisely, the average accuracy of our proposed KT-RCNet outperforms the second-best method by a margin of 5.26%, 3.49%, and 2.25% on miniImageNet, CIFAR100, and CUB200, respectively. Our code is available at https://github.com/YeZiLaiXi/KT-RCNet.git.

Xiaoying Xie, Biao Gong, Yiliang Lv et al.

Most recent works focus on answering first order logical queries to explore the knowledge graph reasoning via multi-hop logic predictions. However, existing reasoning models are limited by the circumscribed logical paradigms of training samples, which leads to a weak generalization of unseen logic. To address these issues, we propose a plug-in module called Logic Diffusion (LoD) to discover unseen queries from surroundings and achieves dynamical equilibrium between different kinds of patterns. The basic idea of LoD is relation diffusion and sampling sub-logic by random walking as well as a special training mechanism called gradient adaption. Besides, LoD is accompanied by a novel loss function to further achieve the robust logical diffusion when facing noisy data in training or testing sets. Extensive experiments on four public datasets demonstrate the superiority of mainstream knowledge graph reasoning models with LoD over state-of-the-art. Moreover, our ablation study proves the general effectiveness of LoD on the noise-rich knowledge graph.

Ye Wang, Yaxiong Wang, Guoshuai Zhao et al.

Few-shot class-incremental learning (FSCIL) aims to incrementally recognize new classes using a few samples while maintaining the performance on previously learned classes. One of the effective methods to solve this challenge is to construct prototypical evolution classifiers. Despite the advancement achieved by most existing methods, the classifier weights are simply initialized using mean features. Because representations for new classes are weak and biased, we argue such a strategy is suboptimal. In this paper, we tackle this issue from two aspects. Firstly, thanks to the development of foundation models, we employ a foundation model, the CLIP, as the network pedestal to provide a general representation for each class. Secondly, to generate a more reliable and comprehensive instance representation, we propose a Knowledge Adapter (KA) module that summarizes the data-specific knowledge from training data and fuses it into the general representation. Additionally, to tune the knowledge learned from the base classes to the upcoming classes, we propose a mechanism of Incremental Pseudo Episode Learning (IPEL) by simulating the actual FSCIL. Taken together, our proposed method, dubbed as Knowledge Adaptation Network (KANet), achieves competitive performance on a wide range of datasets, including CIFAR100, CUB200, and ImageNet-R.

Yimin Deng, Yuqing Fu, Derong Xu et al.

Conversational agents struggle to handle long conversations due to context window limitations. Therefore, memory systems are developed to leverage essential historical information. Existing memory systems typically follow a pipeline of offline memory construction and update, and online retrieval. Despite the flexible online phase, the offline phase remains fixed and task-independent. In this phase, memory construction operates under a predefined workflow and fails to emphasize task relevant information. Meanwhile, memory updates are guided by generic metrics rather than task specific supervision. This leads to a misalignment between offline memory preparation and task requirements, which undermines downstream task performance. To this end, we propose an Adversarial Memory Adaptation mechanism (AMA) that aligns memory construction and update with task objectives by simulating task execution. Specifically, first, a challenger agent generates question answer pairs based on the original dialogues. The constructed memory is then used to answer these questions, simulating downstream inference. Subsequently, an evaluator agent assesses the responses and performs error analysis. Finally, an adapter agent analyzes the error cases and performs dual level updates on both the construction strategy and the content. Through this process, the memory system receives task aware supervision signals in advance during the offline phase, enhancing its adaptability to downstream tasks. AMA can be integrated into various existing memory systems, and extensive experiments on long dialogue benchmark LoCoMo demonstrate its effectiveness.

Yuqing Fu, Yimin Deng, Wanyu Wang et al.

Multi-hop Question Answering (MHQA) aims to answer questions that require multi-step reasoning. It presents two key challenges: generating correct reasoning paths in response to the complex user queries, and accurately retrieving essential knowledge in the face of potential limitations in large language models (LLMs). Existing approaches primarily rely on prompt-based methods to generate reasoning paths, which are further combined with traditional sparse or dense retrieval to produce the final answer. However, the generation of reasoning paths commonly lacks effective control over the generative process, thus leading the reasoning astray. Meanwhile, the retrieval methods over-rely on knowledge matching or similarity scores rather than evaluating the practical utility of the information, resulting in retrieving homogeneous or non-useful information. Therefore, we propose a Structured Entity-Aware Retrieval with Chain-of-Reasoning Navigator framework named SEARCH-R. Specifically, SEARCH-R trains an end-to-end reasoning path navigator, which is able to provide a powerful sub-question decomposer by fine-tuning the Llama3.1-8B model. Moreover, a novel dependency tree-based retrieval is designed to evaluate the informational contribution of the document quantitatively. Extensive experiments on three challenging multi-hop datasets validate the effectiveness of the proposed framework. The code and dataset are available at: https://github.com/Applied-Machine-Learning-Lab/ACL2026_SEARCH-R.

Ye Wang, Yaxiong Wang, Guoshuai Zhao et al.

Continuous Generalized Category Discovery (C-GCD) aims to continually discover novel classes from unlabelled image sets while maintaining performance on old classes. In this paper, we propose a novel learning framework, dubbed Neighborhood Commonality-aware Evolution Network (NCENet) that conquers this task from the perspective of representation learning. Concretely, to learn discriminative representations for novel classes, a Neighborhood Commonality-aware Representation Learning (NCRL) is designed, which exploits local commonalities derived neighborhoods to guide the learning of representational differences between instances of different classes. To maintain the representation ability for old classes, a Bi-level Contrastive Knowledge Distillation (BCKD) module is designed, which leverages contrastive learning to perceive the learning and learned knowledge and conducts knowledge distillation. Extensive experiments conducted on CIFAR10, CIFAR100, and Tiny-ImageNet demonstrate the superior performance of NCENet compared to the previous state-of-the-art method. Particularly, in the last incremental learning session on CIFAR100, the clustering accuracy of NCENet outperforms the second-best method by a margin of 3.09\% on old classes and by a margin of 6.32\% on new classes. Our code will be publicly available at \href{https://github.com/xjtuYW/NCENet.git}{https://github.com/xjtuYW/NCENet.git}. \end{abstract}

Ruizhe Wang, Yeyun Gong, Xiao Liu et al.

The growing computational demands of training large language models (LLMs) necessitate more efficient methods. Quantized training presents a promising solution by enabling low-bit arithmetic operations to reduce these costs. While FP8 precision has demonstrated feasibility, leveraging FP4 remains a challenge due to significant quantization errors and limited representational capacity. This work introduces the first FP4 training framework for LLMs, addressing these challenges with two key innovations: a differentiable quantization estimator for precise weight updates and an outlier clamping and compensation strategy to prevent activation collapse. To ensure stability, the framework integrates a mixed-precision training scheme and vector-wise quantization. Experimental results demonstrate that our FP4 framework achieves accuracy comparable to BF16 and FP8, with minimal degradation, scaling effectively to 13B-parameter LLMs trained on up to 100B tokens. With the emergence of next-generation hardware supporting FP4, our framework sets a foundation for efficient ultra-low precision training.

Yimin Deng, Zhenxi Lin, Yejing Wang et al.

Diagnostic prediction and clinical reasoning are critical tasks in healthcare applications. While Large Language Models (LLMs) have shown strong capabilities in commonsense reasoning, they still struggle with diagnostic reasoning due to limited domain knowledge. Existing approaches often rely on internal model knowledge or static knowledge bases, resulting in knowledge insufficiency and limited adaptability, which hinder their capacity to perform diagnostic reasoning. Moreover, these methods focus solely on the accuracy of final predictions, overlooking alignment with standard clinical reasoning trajectories. To this end, we propose MultiDx, a two-stage diagnostic reasoning framework that performs differential diagnosis by analyzing evidence collected from multiple knowledge sources. Specifically, it first generates suspected diagnoses and reasoning paths by leveraging knowledge from web search, SOAP-formatted case, and clinical case database. Then it integrates multi-perspective evidence through matching, voting, and differential diagnosis to generate the final prediction.~Extensive experiments on two public benchmarks demonstrate the effectiveness of our approach.

Yimin Deng, Yejing Wang, Zhenxi Lin et al.

Large language models have demonstrated strong reasoning capabilities in general knowledge question answering. However, their ability to handle temporal information remains limited. To address this limitation, existing approaches often involve external tools or manual verification and are tailored to specific scenarios, leading to poor generalizability. Moreover, these methods apply a fixed pipeline to all questions, overlooking the fact that different types of temporal questions require distinct reasoning strategies, which leads to unnecessary processing for simple cases and inadequate reasoning for complex ones. To this end, we propose AdapTime, an adaptive temporal reasoning method that dynamically executes reasoning steps based on the input context. Specifically, it involves three temporal reasoning actions: reformulate, rewrite and review, with an LLM planner guiding the reasoning process. AdapTime integrates seamlessly with state-of-the-art LLMs and significantly enhances their temporal reasoning capabilities without relying on external support. Extensive experiments demonstrate the effectiveness of our approach.

Zhenghao Lin, Zihao Tang, Xiao Liu et al.

We introduce Sigma, an efficient large language model specialized for the system domain, empowered by a novel architecture including DiffQKV attention, and pre-trained on our meticulously collected system domain data. DiffQKV attention significantly enhances the inference efficiency of Sigma by optimizing the Query (Q), Key (K), and Value (V) components in the attention mechanism differentially, based on their varying impacts on the model performance and efficiency indicators. Specifically, we (1) conduct extensive experiments that demonstrate the model's varying sensitivity to the compression of K and V components, leading to the development of differentially compressed KV, and (2) propose augmented Q to expand the Q head dimension, which enhances the model's representation capacity with minimal impacts on the inference speed. Rigorous theoretical and empirical analyses reveal that DiffQKV attention significantly enhances efficiency, achieving up to a 33.36% improvement in inference speed over the conventional grouped-query attention (GQA) in long-context scenarios. We pre-train Sigma on 6T tokens from various sources, including 19.5B system domain data that we carefully collect and 1T tokens of synthesized and rewritten data. In general domains, Sigma achieves comparable performance to other state-of-arts models. In the system domain, we introduce the first comprehensive benchmark AIMicius, where Sigma demonstrates remarkable performance across all tasks, significantly outperforming GPT-4 with an absolute improvement up to 52.5%.

Yimin Deng, Yuxia Wu, Guoshuai Zhao et al.

New intent discovery is a crucial capability for task-oriented dialogue systems. Existing methods focus on transferring in-domain (IND) prior knowledge to out-of-domain (OOD) data through pre-training and clustering stages. They either handle the two processes in a pipeline manner, which exhibits a gap between intent representation and clustering process or use typical contrastive clustering that overlooks the potential supervised signals from the whole data. Besides, they often individually deal with open intent discovery or OOD settings. To this end, we propose a Pseudo-Label enhanced Prototypical Contrastive Learning (PLPCL) model for uniformed intent discovery. We iteratively utilize pseudo-labels to explore potential positive/negative samples for contrastive learning and bridge the gap between representation and clustering. To enable better knowledge transfer, we design a prototype learning method integrating the supervised and pseudo signals from IND and OOD samples. In addition, our method has been proven effective in two different settings of discovering new intents. Experiments on three benchmark datasets and two task settings demonstrate the effectiveness of our approach.

Pengbo Guo, Chengxu Liu, Guoshuai Zhao et al.

Eliminating reflections caused by incident light interacting with reflective medium remains an ill-posed problem in the image restoration area. The primary challenge arises from the overlapping of reflection and transmission components in the captured images, which complicates the task of accurately distinguishing and recovering the clean background. Existing approaches typically address reflection removal solely in the image domain, ignoring the spectral property variations of reflected light, which hinders their ability to effectively discern reflections. In this paper, we start with a new perspective on spectral learning, and propose the Spectral Codebook to reconstruct the optical spectrum of the reflection image. The reflections can be effectively distinguished by perceiving the wavelength differences between different light sources in the spectrum. To leverage the reconstructed spectrum, we design two spectral prior refinement modules to re-distribute pixels in the spatial dimension and adaptively enhance the spectral differences along the wavelength dimension. Furthermore, we present the Spectrum-Aware Transformer to jointly recover the transmitted content in spectral and pixel domains. Experimental results on three different reflection benchmarks demonstrate the superiority and generalization ability of our method compared to state-of-the-art models.

Yunqi Mi, Jiakui Shen, Guoshuai Zhao et al.

Extending recommender systems to federated learning (FL) frameworks to protect the privacy of users or platforms while making recommendations has recently gained widespread attention in academia. This is due to the natural coupling of recommender systems and federated learning architectures: the data originates from distributed clients (mostly mobile devices held by users), which are highly related to privacy. In a centralized recommender system (CenRec), the central server collects clients' data, trains the model, and provides the service. Whereas in federated recommender systems (FedRec), the step of data collecting is omitted, and the step of model training is offloaded to each client. The server only aggregates the model and other knowledge, thus avoiding client privacy leakage. Some surveys of federated recommender systems discuss and analyze related work from the perspective of designing FL systems. However, their utility drops by ignoring specific recommendation scenarios' unique characteristics and practical challenges. For example, the statistical heterogeneity issue in cross-domain FedRec originates from the label drift of the data held by different platforms, which is mainly caused by the recommender itself, but not the federated architecture. Therefore, it should focus more on solving specific problems in real-world recommendation scenarios to encourage the deployment FedRec. To this end, this review comprehensively analyzes the coupling of recommender systems and federated learning from the perspective of recommendation researchers and practitioners. We establish a clear link between recommendation scenarios and FL frameworks, systematically analyzing scenario-specific approaches, practical challenges, and potential opportunities. We aim to develop guidance for the real-world deployment of FedRec, bridging the gap between existing research and applications.

Yimin Deng, Yuxia Wu, Yejing Wang et al.

Temporal knowledge graph reasoning aims to predict future events with knowledge of existing facts and plays a key role in various downstream tasks. Previous methods focused on either graph structure learning or semantic reasoning, failing to integrate dual reasoning perspectives to handle different prediction scenarios. Moreover, they lack the capability to capture the inherent differences between historical and non-historical events, which limits their generalization across different temporal contexts. To this end, we propose a Multi-Expert Structural-Semantic Hybrid (MESH) framework that employs three kinds of expert modules to integrate both structural and semantic information, guiding the reasoning process for different events. Extensive experiments on three datasets demonstrate the effectiveness of our approach.

Hao Tang, Guoshuai Zhao, Yuxia Wu et al.

The top-k recommendation is a fundamental task in recommendation systems which is generally learned by comparing positive and negative pairs. The Contrastive Loss (CL) is the key in contrastive learning that has received more attention recently and we find it is well suited for top-k recommendations. However, it is a problem that CL treats the importance of the positive and negative samples as the same. On the one hand, CL faces the imbalance problem of one positive sample and many negative samples. On the other hand, positive items are so few in sparser datasets that their importance should be emphasized. Moreover, the other important issue is that the sparse positive items are still not sufficiently utilized in recommendations. So we propose a new data augmentation method by using multiple positive items (or samples) simultaneously with the CL loss function. Therefore, we propose a Multi-Sample based Contrastive Loss (MSCL) function which solves the two problems by balancing the importance of positive and negative samples and data augmentation. And based on the graph convolution network (GCN) method, experimental results demonstrate the state-of-the-art performance of MSCL. The proposed MSCL is simple and can be applied in many methods. We will release our code on GitHub upon the acceptance.