Lidan Shou

Yuxi Chen, Junming Chen, Chenyu He et al.

As LLM agents evolve, tokens have emerged as the core economic primitives of Agentic AI. However, their exponential consumption introduces severe computational, collaborative, and security bottlenecks. Current surveys remain fragmented across system optimization, architecture design, and trust, lacking a unified framework to evaluate the fundamental trade-off between output quality and economic cost. To bridge this gap, this survey presents the first comprehensive survey of Token Economics. By unifying computer science and economics, we conceptualize tokens as production factors, exchange mediums, and units of account. We synthesize existing literature across a four-dimensional taxonomy: (1) Micro-level (Single Agent): Optimizing budget-constrained factor substitution via neoclassical firm theory. (2) Meso-level (Multi-Agent Systems): Minimizing collaboration friction using transaction cost and principal-agent theories. (3) Macro-level (Agent Ecosystems): Addressing congestion externalities and pricing via mechanism design. (4) Security: Internalizing adversarial threats as endogenous economic constraints. Finally, we outline frontier directions, including differentiable token budgets and dynamic markets, to lay the theoretical foundation for scalable next-generation agent systems.

Yihao Wang, Haoran Xu, Renjie Gu et al.

The large-scale deployment of personalized healthcare agents demands memory mechanisms that are exceptionally precise, safe, and capable of long-term clinical tracking. However, existing benchmarks primarily focus on daily open-domain conversations, failing to capture the high-stakes complexity of real-world medical applications. Motivated by the stringent production requirements of an industry-leading health management agent serving tens of millions of active users, we introduce MedMemoryBench. We develop a human-agent collaborative pipeline to synthesize highly realistic, long-horizon medical trajectories based on clinically grounded, synthetic patient archetypes. This process yields a massive, expertly validated dataset comprising approximately 2,000 sessions and 16,000 interaction turns. Crucially, MedMemoryBench departs from traditional static evaluations by pioneering an "evaluate-while-constructing" streaming assessment protocol, which precisely mirrors dynamic memory accumulation in production environments. Furthermore, we formalize and systematically investigate the critical phenomenon of memory saturation, where sustained information influx actively degrades retrieval and reasoning robustness. Comprehensive benchmarking reveals severe bottlenecks in mainstream architectures, particularly concerning complex medical reasoning and noise resilience. By exposing these fundamental flaws, MedMemoryBench establishes a vital foundation for developing robust, production-ready medical agents.

Cheng Peng, Ke Chen, Lidan Shou et al.

Multi-modal multi-label emotion recognition (MMER) aims to identify relevant emotions from multiple modalities. The challenge of MMER is how to effectively capture discriminative features for multiple labels from heterogeneous data. Recent studies are mainly devoted to exploring various fusion strategies to integrate multi-modal information into a unified representation for all labels. However, such a learning scheme not only overlooks the specificity of each modality but also fails to capture individual discriminative features for different labels. Moreover, dependencies of labels and modalities cannot be effectively modeled. To address these issues, this paper presents ContrAstive feature Reconstruction and AggregaTion (CARAT) for the MMER task. Specifically, we devise a reconstruction-based fusion mechanism to better model fine-grained modality-to-label dependencies by contrastively learning modal-separated and label-specific features. To further exploit the modality complementarity, we introduce a shuffle-based aggregation strategy to enrich co-occurrence collaboration among labels. Experiments on two benchmark datasets CMU-MOSEI and M3ED demonstrate the effectiveness of CARAT over state-of-the-art methods. Code is available at https://github.com/chengzju/CARAT.

Jun Zhang, Jue Wang, Huan Li et al.

Large Language Models (LLMs) have significantly advanced natural language processing with exceptional task generalization capabilities. Low-Rank Adaption (LoRA) offers a cost-effective fine-tuning solution, freezing the original model parameters and training only lightweight, low-rank adapter matrices. However, the memory footprint of LoRA is largely dominated by the original model parameters. To mitigate this, we propose LoRAM, a memory-efficient LoRA training scheme founded on the intuition that many neurons in over-parameterized LLMs have low training utility but are essential for inference. LoRAM presents a unique twist: it trains on a pruned (small) model to obtain pruned low-rank matrices, which are then recovered and utilized with the original (large) model for inference. Additionally, minimal-cost continual pre-training, performed by the model publishers in advance, aligns the knowledge discrepancy between pruned and original models. Our extensive experiments demonstrate the efficacy of LoRAM across various pruning strategies and downstream tasks. For a model with 70 billion parameters, LoRAM enables training on a GPU with only 20G HBM, replacing an A100-80G GPU for LoRA training and 15 GPUs for full fine-tuning. Specifically, QLoRAM implemented by structured pruning combined with 4-bit quantization, for LLaMA-3.1-70B (LLaMA-2-70B), reduces the parameter storage cost that dominates the memory usage in low-rank matrix training by 15.81$\times$ (16.95$\times$), while achieving dominant performance gains over both the original LLaMA-3.1-70B (LLaMA-2-70B) and LoRA-trained LLaMA-3.1-8B (LLaMA-2-13B). Code is available at https://github.com/junzhang-zj/LoRAM.

Yicheng Ji, Jun Zhang, Heming Xia et al.

Video large language models (Vid-LLMs) have shown strong capabilities in understanding video content. However, their reliance on dense video token representations introduces substantial memory and computational overhead in both prefilling and decoding. To mitigate the information loss of recent video token reduction methods and accelerate the decoding stage of Vid-LLMs losslessly, we introduce SpecVLM, a training-free speculative decoding (SD) framework tailored for Vid-LLMs that incorporates staged video token pruning. Building on our novel finding that the draft model's speculation exhibits low sensitivity to video token pruning, SpecVLM prunes up to 90% of video tokens to enable efficient speculation without sacrificing accuracy. To achieve this, we performs a two-stage pruning process: Stage I selects highly informative tokens guided by attention signals from the verifier (target model), while Stage II prunes remaining redundant ones in a spatially uniform manner. Extensive experiments on four video understanding benchmarks demonstrate the effectiveness and robustness of SpecVLM, which achieves up to 2.68$\times$ decoding speedup for LLaVA-OneVision-72B and 2.11$\times$ speedup for Qwen2.5-VL-32B. Code is available at https://github.com/zju-jiyicheng/SpecVLM.

Jun Zhang, Huan Li, Dazhong Rong et al.

Privacy concerns have led to the rise of federated recommender systems (FRS), which can create personalized models across distributed clients. However, FRS is vulnerable to poisoning attacks, where malicious users manipulate gradients to promote their target items intentionally. Existing attacks against FRS have limitations, as they depend on specific models and prior knowledge, restricting their real-world applicability. In our exploration of practical FRS vulnerabilities, we devise a model-agnostic and prior-knowledge-free attack, named PIECK (Popular Item Embedding based Attack). The core module of PIECK is popular item mining, which leverages embedding changes during FRS training to effectively identify the popular items. Built upon the core module, PIECK branches into two diverse solutions: The PIECKIPE solution employs an item popularity enhancement module, which aligns the embeddings of targeted items with the mined popular items to increase item exposure. The PIECKUEA further enhances the robustness of the attack by using a user embedding approximation module, which approximates private user embeddings using mined popular items. Upon identifying PIECK, we evaluate existing federated defense methods and find them ineffective against PIECK, as poisonous gradients inevitably overwhelm the cold target items. We then propose a novel defense method by introducing two regularization terms during user training, which constrain item popularity enhancement and user embedding approximation while preserving FRS performance. We evaluate PIECK and its defense across two base models, three real datasets, four top-tier attacks, and six general defense methods, affirming the efficacy of both PIECK and its defense.

Yongqin Xu, Huan Li, Ke Chen et al.

Schema matching (SM) and entity matching (EM) tasks are crucial for data integration. While large language models (LLMs) have shown promising results in these tasks, they suffer from hallucinations and confusion about task instructions. This study presents the Knowledge-Compliant Matching Framework (KcMF), an LLM-based approach that addresses these issues without the need for domain-specific fine-tuning. KcMF employs a once-and-for-all pseudo-code-based task decomposition strategy to adopt natural language statements that guide LLM reasoning and reduce confusion across various task types. We also propose two mechanisms, Dataset as Knowledge (DaK) and Example as Knowledge (EaK), to build domain knowledge sets when unstructured domain knowledge is lacking. Moreover, we introduce a result-ensemble strategy to leverage multiple knowledge sources and suppress badly formatted outputs. Extensive evaluations confirm that KcMF clearly enhances five LLM backbones in both SM and EM tasks while outperforming the non-LLM competitors by an average F1-score of 17.93%.

Yuxin Zhou, Zheng Li, Jun Zhang et al.

With the widespread adoption of Mixture-of-Experts (MoE) models, there is a growing demand for efficient inference on memory-constrained devices. While offloading expert parameters to CPU memory and loading activated experts on demand has emerged as a potential solution, the large size of activated experts overburdens the limited PCIe bandwidth, hindering the effectiveness in latency-sensitive scenarios. To mitigate this, we propose FloE, an on-the-fly MoE inference system on memory-constrained GPUs. FloE is built on the insight that there exists substantial untapped redundancy within sparsely activated experts. It employs various compression techniques on the expert's internal parameter matrices to reduce the data movement load, combined with low-cost sparse prediction, achieving perceptible inference acceleration in wall-clock time on resource-constrained devices. Empirically, FloE achieves a 9.3x compression of parameters per expert in Mixtral-8x7B; enables deployment on a GPU with only 11GB VRAM, reducing the memory footprint by up to 8.5x; and delivers a 48.7x inference speedup compared to DeepSpeed-MII on a single GeForce RTX 3090 - all with only a 4.4$\%$ - 7.6$\%$ average performance degradation.

Hong Lin, Lidan Shou, Ke Chen et al.

Federated learning (FL) is a promising approach for learning a model from data distributed on massive clients without exposing data privacy. It works effectively in the ideal federation where clients share homogeneous data distribution and learning behavior. However, FL may fail to function appropriately when the federation is not ideal, amid an unhealthy state called Negative Federated Learning (NFL), in which most clients gain no benefit from participating in FL. Many studies have tried to address NFL. However, their solutions either (1) predetermine to prevent NFL in the entire learning life-cycle or (2) tackle NFL in the aftermath of numerous learning rounds. Thus, they either (1) indiscriminately incur extra costs even if FL can perform well without such costs or (2) waste numerous learning rounds. Additionally, none of the previous work takes into account the clients who may be unwilling/unable to follow the proposed NFL solutions when using those solutions to upgrade an FL system in use. This paper introduces FL-GUARD, a holistic framework that can be employed on any FL system for tackling NFL in a run-time paradigm. That is, to dynamically detect NFL at the early stage (tens of rounds) of learning and then to activate recovery measures when necessary. Specifically, we devise a cost-effective NFL detection mechanism, which relies on an estimation of performance gain on clients. Only when NFL is detected, we activate the NFL recovery process, in which each client learns in parallel an adapted model when training the global model. Extensive experiment results confirm the effectiveness of FL-GUARD in detecting NFL and recovering from NFL to a healthy learning state. We also show that FL-GUARD is compatible with previous NFL solutions and robust against clients unwilling/unable to take any recovery measures.

Jun Zhang, Jue Wang, Huan Li et al.

The significant computational demands of pretrained language models (PLMs), which often require dedicated hardware, present a substantial challenge in serving them efficiently, especially in multi-tenant environments. To address this, we introduce HMI, a Hierarchical knowledge management-based Multi-tenant Inference system, designed to manage tenants with distinct PLMs resource-efficiently. Our approach is three-fold: Firstly, we categorize PLM knowledge into general, domain-specific, and task-specific. Leveraging insights on knowledge acquisition across different model layers, we construct hierarchical PLMs (hPLMs) by extracting and storing knowledge at different levels, significantly reducing GPU memory usage per tenant. Secondly, we establish hierarchical knowledge management for hPLMs generated by various tenants in HMI. We manage domain-specific knowledge with acceptable storage increases by constructing and updating domain-specific knowledge trees based on frequency. We manage task-specific knowledge within limited GPU memory through parameter swapping. Finally, we propose system optimizations to enhance resource utilization and inference throughput. These include fine-grained pipelining via hierarchical knowledge prefetching to overlap CPU and I/O operations with GPU computations, and optimizing parallel implementations with batched matrix multiplications. Our experimental results demonstrate that the proposed HMI can efficiently serve up to 10,000 hPLMs (hBERTs and hGPTs) on a single GPU, with only a negligible compromise in accuracy.

Jun Zhang, Jue Wang, Huan Li et al.

Active learning (AL) reduces human annotation costs for machine learning systems by strategically selecting the most informative unlabeled data for annotation, but performing it individually may still be insufficient due to restricted data diversity and annotation budget. Federated Active Learning (FAL) addresses this by facilitating collaborative data selection and model training, while preserving the confidentiality of raw data samples. Yet, existing FAL methods fail to account for the heterogeneity of data distribution across clients and the associated fluctuations in global and local model parameters, adversely affecting model accuracy. To overcome these challenges, we propose CHASe (Client Heterogeneity-Aware Data Selection), specifically designed for FAL. CHASe focuses on identifying those unlabeled samples with high epistemic variations (EVs), which notably oscillate around the decision boundaries during training. To achieve both effectiveness and efficiency, \model{} encompasses techniques for 1) tracking EVs by analyzing inference inconsistencies across training epochs, 2) calibrating decision boundaries of inaccurate models with a new alignment loss, and 3) enhancing data selection efficiency via a data freeze and awaken mechanism with subset sampling. Experiments show that CHASe surpasses various established baselines in terms of effectiveness and efficiency, validated across diverse datasets, model complexities, and heterogeneous federation settings.

Jun Zhang, Jue Wang, Huan Li et al.

We present a novel inference scheme, self-speculative decoding, for accelerating Large Language Models (LLMs) without the need for an auxiliary model. This approach is characterized by a two-stage process: drafting and verification. The drafting stage generates draft tokens at a slightly lower quality but more quickly, which is achieved by selectively skipping certain intermediate layers during drafting. Subsequently, the verification stage employs the original LLM to validate those draft output tokens in one forward pass. This process ensures the final output remains identical to that produced by the unaltered LLM. Moreover, the proposed method requires no additional neural network training and no extra memory footprint, making it a plug-and-play and cost-effective solution for inference acceleration. Benchmarks with LLaMA-2 and its variants demonstrated a speedup up to 1.99$\times$.

Jue Wang, Ke Chen, Lidan Shou et al.

This paper addresses the problem of key phrase extraction from sentences. Existing state-of-the-art supervised methods require large amounts of annotated data to achieve good performance and generalization. Collecting labeled data is, however, often expensive. In this paper, we redefine the problem as question-answer extraction, and present SAMIE: Self-Asking Model for Information Ixtraction, a semi-supervised model which dually learns to ask and to answer questions by itself. Briefly, given a sentence $s$ and an answer $a$, the model needs to choose the most appropriate question $\hat q$; meanwhile, for the given sentence $s$ and same question $\hat q$ selected in the previous step, the model will predict an answer $\hat a$. The model can support few-shot learning with very limited supervision. It can also be used to perform clustering analysis when no supervision is provided. Experimental results show that the proposed method outperforms typical supervised methods especially when given little labeled data.