Process Reward Models That Think

Step-by-step verifiers -- also known as process reward models (PRMs) -- are a key ingredient for test-time scaling. PRMs require step-level supervision, making them expensive to train. This work aims to build data-efficient PRMs as verbalized step-wise reward models that verify every step in the solution by generating a verification chain-of-thought (CoT). We propose ThinkPRM, a long CoT verifier fine-tuned on orders of magnitude fewer process labels than those required by discriminative PRMs. Our approach capitalizes on the inherent reasoning abilities of long CoT models, and outperforms LLM-as-a-Judge and discriminative verifiers -- using only 1% of the process labels in PRM800K -- across several challenging benchmarks. Specifically, ThinkPRM beats the baselines on ProcessBench, MATH-500, and AIME '24 under best-of-N selection and reward-guided search. In an out-of-domain evaluation on a subset of GPQA-Diamond and LiveCodeBench, our PRM surpasses discriminative verifiers trained on the full PRM800K by 8% and 4.5%, respectively. Lastly, under the same token budget, ThinkPRM scales up verification compute more effectively compared to LLM-as-a-Judge, outperforming it by 7.2% on a subset of ProcessBench. Our work highlights the value of generative, long CoT PRMs that can scale test-time compute for verification while requiring minimal supervision for training. Our code, data, and models will be released at https://github.com/mukhal/thinkprm.

IsoSEL: Isometric Structural Entropy Learning for Deep Graph Clustering in Hyperbolic Space

Graph clustering is a longstanding topic in machine learning. In recent years, deep learning methods have achieved encouraging results, but they still require predefined cluster numbers K, and typically struggle with imbalanced graphs, especially in identifying minority clusters. The limitations motivate us to study a challenging yet practical problem: deep graph clustering without K considering the imbalance in reality. We approach this problem from a fresh perspective of information theory (i.e., structural information). In the literature, structural information has rarely been touched in deep clustering, and the classic definition falls short in its discrete formulation, neglecting node attributes and exhibiting prohibitive complexity. In this paper, we first establish a new Differentiable Structural Information, generalizing the discrete formalism to continuous realm, so that the optimal partitioning tree, revealing the cluster structure, can be created by the gradient backpropagation. Theoretically, we demonstrate its capability in clustering without requiring K and identifying the minority clusters in imbalanced graphs, while reducing the time complexity to O(N) w.r.t. the number of nodes. Subsequently, we present a novel IsoSEL framework for deep graph clustering, where we design a hyperbolic neural network to learn the partitioning tree in the Lorentz model of hyperbolic space, and further conduct Lorentz Tree Contrastive Learning with isometric augmentation. As a result, the partitioning tree incorporates node attributes via mutual information maximization, while the cluster assignment is refined by the proposed tree contrastive learning. Extensive experiments on five benchmark datasets show the IsoSEL outperforms 14 recent baselines by an average of +1.3% in NMI.

Certainly Bot Or Not? Trustworthy Social Bot Detection via Robust Multi-Modal Neural Processes

Social bot detection is crucial for mitigating misinformation, online manipulation, and coordinated inauthentic behavior. While existing neural network-based detectors perform well on benchmarks, they struggle with generalization due to distribution shifts across datasets and frequently produce overconfident predictions for out-of-distribution accounts beyond the training data. To address this, we introduce a novel Uncertainty Estimation for Social Bot Detection (UESBD) framework, which quantifies the predictive uncertainty of detectors beyond mere classification. For this task, we propose Robust Multi-modal Neural Processes (RMNP), which aims to enhance the robustness of multi-modal neural processes to modality inconsistencies caused by social bot camouflage. RMNP first learns unimodal representations through modality-specific encoders. Then, unimodal attentive neural processes are employed to encode the Gaussian distribution of unimodal latent variables. Furthermore, to avoid social bots stealing human features to camouflage themselves thus causing certain modalities to provide conflictive information, we introduce an evidential gating network to explicitly model the reliability of modalities. The joint latent distribution is learned through the generalized product of experts, which takes the reliability of each modality into consideration during fusion. The final prediction is obtained through Monte Carlo sampling of the joint latent distribution followed by a decoder. Experiments on three real-world benchmarks show the effectiveness of RMNP in classification and uncertainty estimation, as well as its robustness to modality conflicts.

MRCEval: A Comprehensive, Challenging and Accessible Machine Reading Comprehension Benchmark

Machine Reading Comprehension (MRC) is an essential task in evaluating natural language understanding. Existing MRC datasets primarily assess specific aspects of reading comprehension (RC), lacking a comprehensive MRC benchmark. To fill this gap, we first introduce a novel taxonomy that categorizes the key capabilities required for RC. Based on this taxonomy, we construct MRCEval, an MRC benchmark that leverages advanced Large Language Models (LLMs) as both sample generators and selection judges. MRCEval is a comprehensive, challenging and accessible benchmark designed to assess the RC capabilities of LLMs thoroughly, covering 13 distinct RC skills with a total of 2.1K high-quality multi-choice questions. We perform an extensive evaluation of 28 widely used open-source and proprietary models, highlighting that MRC continues to present significant challenges even in the era of LLMs.

Training Large Recommendation Models via Graph-Language Token Alignment

Recommender systems (RS) have become essential tools for helping users efficiently navigate the overwhelming amount of information on e-commerce and social platforms. However, traditional RS relying on Collaborative Filtering (CF) struggles to integrate the rich semantic information from textual data. Meanwhile, large language models (LLMs) have shown promising results in natural language processing, but directly using LLMs for recommendation introduces challenges, such as ambiguity in generating item predictions and inefficiencies in scalability. In this paper, we propose a novel framework to train Large Recommendation models via Graph-Language Token Alignment. By aligning item and user nodes from the interaction graph with pretrained LLM tokens, GLTA effectively leverages the reasoning abilities of LLMs. Furthermore, we introduce Graph-Language Logits Matching (GLLM) to optimize token alignment for end-to-end item prediction, eliminating ambiguity in the free-form text as recommendation results. Extensive experiments on three benchmark datasets demonstrate the effectiveness of GLTA, with ablation studies validating each component.

Agentic Reward Modeling: Integrating Human Preferences with Verifiable Correctness Signals for Reliable Reward Systems

Reward models (RMs) are crucial for the training and inference-time scaling up of large language models (LLMs). However, existing reward models primarily focus on human preferences, neglecting verifiable correctness signals which have shown strong potential in training LLMs. In this paper, we propose agentic reward modeling, a reward system that combines reward models with verifiable correctness signals from different aspects to provide reliable rewards. We empirically implement a reward agent, named RewardAgent, that combines human preference rewards with two verifiable signals: factuality and instruction following, to provide more reliable rewards. We conduct comprehensive experiments on existing reward model benchmarks and inference time best-of-n searches on real-world downstream tasks. RewardAgent significantly outperforms vanilla reward models, demonstrating its effectiveness. We further construct training preference pairs using RewardAgent and train an LLM with the DPO objective, achieving superior performance on various NLP benchmarks compared to conventional reward models. Our codes are publicly released to facilitate further research (https://github.com/THU-KEG/Agentic-Reward-Modeling).

SEA: Low-Resource Safety Alignment for Multimodal Large Language Models via Synthetic Embeddings

Multimodal Large Language Models (MLLMs) have serious security vulnerabilities.While safety alignment using multimodal datasets consisting of text and data of additional modalities can effectively enhance MLLM's security, it is costly to construct these datasets. Existing low-resource security alignment methods, including textual alignment, have been found to struggle with the security risks posed by additional modalities. To address this, we propose Synthetic Embedding augmented safety Alignment (SEA), which optimizes embeddings of additional modality through gradient updates to expand textual datasets. This enables multimodal safety alignment training even when only textual data is available. Extensive experiments on image, video, and audio-based MLLMs demonstrate that SEA can synthesize a high-quality embedding on a single RTX3090 GPU within 24 seconds. SEA significantly improves the security of MLLMs when faced with threats from additional modalities. To assess the security risks introduced by video and audio, we also introduced a new benchmark called VA-SafetyBench. High attack success rates across multiple MLLMs validate its challenge. Our code and data will be available at https://github.com/ZeroNLP/SEA.

The Best Instruction-Tuning Data are Those That Fit

High-quality supervised fine-tuning (SFT) data are crucial for eliciting strong capabilities from pretrained large language models (LLMs). Typically, instructions are paired with multiple responses sampled from other LLMs, which are often out of the distribution of the target model to be fine-tuned. This, at scale, can lead to diminishing returns and even hurt the models' performance and robustness. We propose **GRAPE**, a novel SFT framework that accounts for the unique characteristics of the target model. For each instruction, it gathers responses from various LLMs and selects the one with the highest probability measured by the target model, indicating that it aligns most closely with the target model's pretrained distribution; it then proceeds with standard SFT training. We first evaluate GRAPE with a controlled experiment, where we sample various solutions for each question in UltraInteract from multiple models and fine-tune commonly used LMs like LLaMA3.1-8B, Mistral-7B, and Qwen2.5-7B on GRAPE-selected data. GRAPE significantly outperforms strong baselines, including distilling from the strongest model with an absolute gain of up to 13.8%, averaged across benchmarks, and training on 3x more data with a maximum performance improvement of 17.3%. GRAPE's strong performance generalizes to realistic settings. We experiment with the post-training data used for Tulu3 and Olmo-2. GRAPE outperforms strong baselines trained on 4.5 times more data by 6.1% and a state-of-the-art data selection approach by 3% on average performance. Remarkably, using 1/3 of the data and half the number of epochs, GRAPE enables LLaMA3.1-8B to surpass the performance of Tulu3-SFT by 3.5%.

RiemannGFM: Learning a Graph Foundation Model from Riemannian Geometry

The foundation model has heralded a new era in artificial intelligence, pretraining a single model to offer cross-domain transferability on different datasets. Graph neural networks excel at learning graph data, the omnipresent non-Euclidean structure, but often lack the generalization capacity. Hence, graph foundation model is drawing increasing attention, and recent efforts have been made to leverage Large Language Models. On the one hand, existing studies primarily focus on text-attributed graphs, while a wider range of real graphs do not contain fruitful textual attributes. On the other hand, the sequential graph description tailored for the Large Language Model neglects the structural complexity, which is a predominant characteristic of the graph. Such limitations motivate an important question: Can we go beyond Large Language Models, and pretrain a universal model to learn the structural knowledge for any graph? The answer in the language or vision domain is a shared vocabulary. We observe the fact that there also exist shared substructures underlying graph domain, and thereby open a new opportunity of graph foundation model with structural vocabulary. The key innovation is the discovery of a simple yet effective structural vocabulary of trees and cycles, and we explore its inherent connection to Riemannian geometry. Herein, we present a universal pretraining model, RiemannGFM. Concretely, we first construct a novel product bundle to incorporate the diverse geometries of the vocabulary. Then, on this constructed space, we stack Riemannian layers where the structural vocabulary, regardless of specific graph, is learned in Riemannian manifold offering cross-domain transferability. Extensive experiments show the effectiveness of RiemannGFM on a diversity of real graphs.

Pioneer: Physics-informed Riemannian Graph ODE for Entropy-increasing Dynamics

Dynamic interacting system modeling is important for understanding and simulating real world systems. The system is typically described as a graph, where multiple objects dynamically interact with each other and evolve over time. In recent years, graph Ordinary Differential Equations (ODE) receive increasing research attentions. While achieving encouraging results, existing solutions prioritize the traditional Euclidean space, and neglect the intrinsic geometry of the system and physics laws, e.g., the principle of entropy increasing. The limitations above motivate us to rethink the system dynamics from a fresh perspective of Riemannian geometry, and pose a more realistic problem of physics-informed dynamic system modeling, considering the underlying geometry and physics law for the first time. In this paper, we present a novel physics-informed Riemannian graph ODE for a wide range of entropy-increasing dynamic systems (termed as Pioneer). In particular, we formulate a differential system on the Riemannian manifold, where a manifold-valued graph ODE is governed by the proposed constrained Ricci flow, and a manifold preserving Gyro-transform aware of system geometry. Theoretically, we report the provable entropy non-decreasing of our formulation, obeying the physics laws. Empirical results show the superiority of Pioneer on real datasets.