HumanVBench: Exploring Human-Centric Video Understanding Capabilities of MLLMs with Synthetic Benchmark Data

In the domain of Multimodal Large Language Models (MLLMs), achieving human-centric video understanding remains a formidable challenge. Existing benchmarks primarily emphasize object and action recognition, often neglecting the intricate nuances of human emotions, behaviors, and speech visual alignment within video content. We present HumanVBench, an innovative benchmark meticulously crafted to bridge these gaps in the evaluation of video MLLMs. HumanVBench comprises 17 carefully designed tasks that explore two primary dimensions: inner emotion and outer manifestations, spanning static and dynamic, basic and complex, as well as single-modal and cross-modal aspects. With two advanced automated pipelines for video annotation and distractor-included QA generation, HumanVBench utilizes diverse state-of-the-art (SOTA) techniques to streamline benchmark data synthesis and quality assessment, minimizing human annotation dependency tailored to human-centric multimodal attributes. A comprehensive evaluation across 16 SOTA video MLLMs reveals notable limitations in current performance, especially in cross-modal and temporal alignment, underscoring the necessity for further refinement toward achieving more human-like understanding. HumanVBench is open-sourced to facilitate future advancements and real-world applications in video MLLMs.

A Simple and Provable Scaling Law for the Test-Time Compute of Large Language Models

We propose a general two-stage algorithm that enjoys a provable scaling law for the test-time compute of large language models (LLMs). Given an input problem, the proposed algorithm first generates $N$ candidate solutions, and then chooses the best one via a multiple-round knockout tournament where each pair of candidates are compared for $K$ times and only the winners move on to the next round. In a minimalistic implementation, both stages can be executed with a black-box LLM alone and nothing else (e.g., no external verifier or reward model), and a total of $N \times (K + 1)$ highly parallelizable LLM calls are needed for solving an input problem. Assuming that a generated candidate solution is correct with probability $p_{\text{gen}} > 0$ and a comparison between a pair of correct and incorrect solutions identifies the right winner with probability $p_{\text{comp}} > 0.5$ (i.e., better than a random guess), we prove theoretically that the failure probability of the proposed algorithm decays to zero exponentially with respect to $N$ and $K$: $$\mathbb{P}(\text{final output is incorrect}) \le (1 - p_{\text{gen}})^N + \lceil \log_2 N \rceil e^{-2 K (p_{\text{comp}} - 0.5)^2}.$$ Our empirical results with the challenging MMLU-Pro benchmark validate the technical assumptions, as well as the efficacy of the proposed algorithm and the gains from scaling up its test-time compute.

What is Wrong with Perplexity for Long-context Language Modeling?

Handling long-context inputs is crucial for large language models (LLMs) in tasks such as extended conversations, document summarization, and many-shot in-context learning. While recent approaches have extended the context windows of LLMs and employed perplexity (PPL) as a standard evaluation metric, PPL has proven unreliable for assessing long-context capabilities. The underlying cause of this limitation has remained unclear. In this work, we provide a comprehensive explanation for this issue. We find that PPL overlooks key tokens, which are essential for long-context understanding, by averaging across all tokens and thereby obscuring the true performance of models in long-context scenarios. To address this, we propose \textbf{LongPPL}, a novel metric that focuses on key tokens by employing a long-short context contrastive method to identify them. Our experiments demonstrate that LongPPL strongly correlates with performance on various long-context benchmarks (e.g., Pearson correlation of -0.96), significantly outperforming traditional PPL in predictive accuracy. Additionally, we introduce \textbf{LongCE} (Long-context Cross-Entropy) loss, a re-weighting strategy for fine-tuning that prioritizes key tokens, leading to consistent improvements across diverse benchmarks. In summary, these contributions offer deeper insights into the limitations of PPL and present effective solutions for accurately evaluating and enhancing the long-context capabilities of LLMs. Code is available at https://github.com/PKU-ML/LongPPL.

Make LLMs better zero-shot reasoners: Structure-orientated autonomous reasoning

Zero-shot reasoning methods with Large Language Models (LLMs) offer significant advantages including great generalization to novel tasks and reduced dependency on human-crafted examples. However, the current zero-shot methods still have limitations in complex tasks, e.g., answering questions that require multi-step reasoning. In this paper, we address this limitation by introducing a novel structure-oriented analysis method to help LLMs better understand the question and guide the problem-solving process of LLMs. We first demonstrate how the existing reasoning strategies, Chain-of-Thought and ReAct, can benefit from our structure-oriented analysis. In addition to empirical investigations, we leverage the probabilistic graphical model to theoretically explain why our structure-oriented analysis can improve the LLM reasoning process. To further improve the reliability in complex question-answering tasks, we propose a multi-agent reasoning system, Structure-oriented Autonomous Reasoning Agents (SARA), that can better enforce the reasoning process following our structure-oriented analysis by refinement techniques and is equipped with external knowledge retrieval capability to reduce factual errors. Extensive experiments verify the effectiveness of the proposed reasoning system. Surprisingly, in some cases, the system even surpasses few-shot methods. Finally, the system not only improves reasoning accuracy in complex tasks but also demonstrates robustness against potential attacks that corrupt the reasoning process.

$α$-DPO: Adaptive Reward Margin is What Direct Preference Optimization Needs

Aligning large language models (LLMs) with human values and intentions is crucial for their utility, honesty, and safety. Reinforcement learning from human feedback (RLHF) is a popular approach to achieve this alignment, but it faces challenges in computational efficiency and training stability. Recent methods like Direct Preference Optimization (DPO) and Simple Preference Optimization (SimPO) have proposed offline alternatives to RLHF, simplifying the process by reparameterizing the reward function. However, DPO depends on a potentially suboptimal reference model, and SimPO's assumption of a fixed target reward margin may lead to suboptimal decisions in diverse data settings. In this work, we propose $\alpha$-DPO, an adaptive preference optimization algorithm designed to address these limitations by introducing a dynamic reward margin. Specifically, $\alpha$-DPO employs an adaptive preference distribution, balancing the policy model and the reference model to achieve personalized reward margins. We provide theoretical guarantees for $\alpha$-DPO, demonstrating its effectiveness as a surrogate optimization objective and its ability to balance alignment and diversity through KL divergence control. Empirical evaluations on AlpacaEval 2 and Arena-Hard show that $\alpha$-DPO consistently outperforms DPO and SimPO across various model settings, establishing it as a robust approach for fine-tuning LLMs. Our method achieves significant improvements in win rates, highlighting its potential as a powerful tool for LLM alignment. The code is available at https://github.com/junkangwu/alpha-DPO

Dynamic and Textual Graph Generation Via Large-Scale LLM-based Agent Simulation

Graph generation is a fundamental task that has been extensively studied in social, technological, and scientific analysis. For modeling the dynamic graph evolution process, traditional rule-based methods struggle to capture community structures within graphs, while deep learning methods only focus on fitting training graphs. This limits existing graph generators to producing graphs that adhere to predefined rules or closely resemble training datasets, achieving poor performance in dynamic graph generation. Given that graphs are abstract representations arising from pairwise interactions in human activities, a realistic simulation of human-wise interaction could provide deeper insights into the graph evolution mechanism. With the increasing recognition of large language models (LLMs) in simulating human behavior, we introduce GraphAgent-Generator (GAG), a novel simulation-based framework for dynamic graph generation. Without training or fine-tuning process of LLM, our framework effectively replicates seven macro-level structural characteristics in established network science theories while surpassing existing baselines in graph expansion tasks by 31\% on specific evaluation metrics. Through node classification task, we validate GAG effectively preserves characteristics of real-world network for node-wise textual features in generated text-rich graph. Furthermore, by incorporating parallel acceleration, GAG supports generating graphs with up to nearly 100,000 nodes or 10 million edges through large-scale LLM-based agent simulation, with a minimum speed-up of 90.4\%. The source code is available at https://anonymous.4open.science/r/GraphAgent-2206.

GenSim: A General Social Simulation Platform with Large Language Model based Agents

With the rapid advancement of large language models (LLMs), recent years have witnessed many promising studies on leveraging LLM-based agents to simulate human social behavior. While prior work has demonstrated significant potential across various domains, much of it has focused on specific scenarios involving a limited number of agents and has lacked the ability to adapt when errors occur during simulation. To overcome these limitations, we propose a novel LLM-agent-based simulation platform called \textit{GenSim}, which: (1) \textbf{Abstracts a set of general functions} to simplify the simulation of customized social scenarios; (2) \textbf{Supports one hundred thousand agents} to better simulate large-scale populations in real-world contexts; (3) \textbf{Incorporates error-correction mechanisms} to ensure more reliable and long-term simulations. To evaluate our platform, we assess both the efficiency of large-scale agent simulations and the effectiveness of the error-correction mechanisms. To our knowledge, GenSim represents an initial step toward a general, large-scale, and correctable social simulation platform based on LLM agents, promising to further advance the field of social science.

Agent-Oriented Planning in Multi-Agent Systems

Through the collaboration of multiple agents possessing diverse expertise and tools, multi-agent systems achieve impressive progress in solving real-world problems. Given the user queries, the meta-agents, serving as the brain within these systems, are required to decompose the queries into multiple sub-tasks that can be allocated to suitable agents capable of solving them, so-called agent-oriented planning. In this study, we identify three critical design principles of agent-oriented planning, including solvability, completeness, and non-redundancy, to ensure that each sub-task is effectively resolved, leading to satisfactory responses to the original queries. These principles further inspire us to propose a novel framework for agent-oriented planning in multi-agent systems, leveraging a fast task decomposition and allocation process followed by an effective and efficient evaluation via a reward model. During the planning process, the meta-agent is also responsible for evaluating the performance of the expert agents, making timely adjustments to the sub-tasks and scheduling as necessary. Besides, we integrate a feedback loop into the proposed framework to further enhance the effectiveness and robustness of such a problem-solving process. Extensive experiments demonstrate the advancement of the proposed framework in solving real-world problems compared to both single-agent systems and existing planning strategies for multi-agent systems.

Exploring Selective Layer Fine-Tuning in Federated Learning

Federated learning (FL) has emerged as a promising paradigm for fine-tuning foundation models using distributed data in a privacy-preserving manner. Under limited computational resources, clients often find it more practical to fine-tune a selected subset of layers, rather than the entire model, based on their task-specific data. In this study, we provide a thorough theoretical exploration of selective layer fine-tuning in FL, emphasizing a flexible approach that allows the clients to adjust their selected layers according to their local data and resources. We theoretically demonstrate that the layer selection strategy has a significant impact on model convergence in two critical aspects: the importance of selected layers and the heterogeneous choices across clients. Drawing from these insights, we further propose a strategic layer selection method that utilizes local gradients and regulates layer selections across clients. The extensive experiments on both image and text datasets demonstrate the effectiveness of the proposed strategy compared with several baselines, highlighting its advances in identifying critical layers that adapt to the client heterogeneity and training dynamics in FL.

Understanding Byzantine Robustness in Federated Learning with A Black-box Server

Federated learning (FL) becomes vulnerable to Byzantine attacks where some of participators tend to damage the utility or discourage the convergence of the learned model via sending their malicious model updates. Previous works propose to apply robust rules to aggregate updates from participators against different types of Byzantine attacks, while at the same time, attackers can further design advanced Byzantine attack algorithms targeting specific aggregation rule when it is known. In practice, FL systems can involve a black-box server that makes the adopted aggregation rule inaccessible to participants, which can naturally defend or weaken some Byzantine attacks. In this paper, we provide an in-depth understanding on the Byzantine robustness of the FL system with a black-box server. Our investigation demonstrates the improved Byzantine robustness of a black-box server employing a dynamic defense strategy. We provide both empirical evidence and theoretical analysis to reveal that the black-box server can mitigate the worst-case attack impact from a maximum level to an expectation level, which is attributed to the inherent inaccessibility and randomness offered by a black-box server.The source code is available at https://github.com/alibaba/FederatedScope/tree/Byzantine_attack_defense to promote further research in the community.