Song Yu

ByteDance Seed, Jiaze Chen, Tiantian Fan et al. · bytedance

We introduce Seed1.5-Thinking, capable of reasoning through thinking before responding, resulting in improved performance on a wide range of benchmarks. Seed1.5-Thinking achieves 86.7 on AIME 2024, 55.0 on Codeforces and 77.3 on GPQA, demonstrating excellent reasoning abilities in STEM and coding. Beyond reasoning tasks, the method demonstrates notable generalization across diverse domains. For instance, it surpasses DeepSeek R1 by 8% in win rate on non-reasoning tasks, indicating its broader applicability. Compared to other state-of-the-art reasoning models, Seed1.5-Thinking is a Mixture-of-Experts (MoE) model with a relatively small size, featuring 20B activated and 200B total parameters. As part of our effort to assess generalized reasoning, we develop two internal benchmarks, BeyondAIME and Codeforces, both of which will be publicly released to support future research. Model trial link: https://www.volcengine.com/experience/ark.

Song Yu, Yang Hu, Haokang Ding et al.

Camouflaged Object Detection is challenging due to the high degree of similarity between camouflaged objects and their surrounding backgrounds. Current COD methods mainly rely on edge extraction in the spatial domain and local pixel-level information, neglecting the importance of global structural features. Additionally, they fail to effectively leverage the importance of phase spectrum information within frequency domain features. To this end, we propose a COD framework BASFNet based on boundary-aware frequency domain and spatial domain fusion.This method uses dual guided integration of frequency domain and spatial domain features. A phase-spectrum-based frequency-enhanced edge exploration module (FEEM) and a spatial core segmentation module (SCSM) are introduced to jointly capture the boundary and object features of camouflaged objects. These features are then effectively integrated through a spatial-frequency fusion interaction module (SFFIM). Furthermore, the boundary detection is further optimized through an boundary-aware training strategy. BASFNet outperforms existing state-of-the-art methods on three benchmark datasets, validating the effectiveness of the fusion of frequency and spatial domain information in COD tasks.

Song Yu, Xiaofei Xu, Ke Deng et al.

Large language models (LLMs) face persistent challenges when handling long-context tasks, most notably the lost in the middle issue, where information located in the middle of a long input tends to be underutilized. Some existing methods that reduce input have the risk of discarding key information, while others that extend context windows often lead to attention dispersion. To address these limitations, we propose Tree of Agents (TOA), a multi-agent reasoning framework that segments the input into chunks processed by independent agents. Each agent generates its local cognition, then agents dynamically exchange information for collaborative reasoning along tree-structured paths. TOA enables agents to probe different reasoning orders for multi-perspective understanding, effectively mitigating position bias and reducing hallucinations. To improve processing efficiency, we incorporate prefix-hash caching and adaptive pruning strategies, achieving significant performance improvements with comparable API overhead. Experiments show that TOA, powered by compact LLaMA3.1-8B, significantly outperforms multiple baselines and demonstrates comparable performance to the latest and much larger commercial models, such as Gemini1.5-pro, on various long-context tasks. Code is available at https://github.com/Aireduce952/Tree-of-Agents.

Song Yu, Li Li, Wenwen Zhao et al.

Reinforcement learning with verifiable rewards (RLVR), particularly Group Relative Policy Optimization (GRPO), has advanced LLM reasoning. However, GRPO suffers from three credit assignment failures: uniform token-level granularity that ignores heterogeneous informational value, uniform polarity that penalizes correct steps and rewards incorrect ones, and zero-variance collapse that erases outcome-driven gradients. We systematically quantify these failures, revealing highly non-uniform token informativeness, widespread step-level polarity misalignment, and substantial training waste. To address these limitations, we propose Entropy-Progress Aligned GRPO (EP-GRPO), a framework that mines the model's intrinsic information flow for dense, self-supervised guidance. EP-GRPO integrates entropy-gated modulation to prioritize high entropy decision pivots, implicit process signals from policy divergence anchored to outcome advantages for directional token-level feedback without external reward models, and cumulative entropy mapping that enables progress-aligned advantage normalization, naturally maintaining gradient flow under zero reward variance. Extensive experiments on mathematical reasoning benchmarks demonstrate that EP-GRPO achieves superior accuracy and efficiency compared to GRPO and its variants. The code will be available.

Song Yu, Li Li

Reinforcement learning from verifiable rewards (RLVR) has significantly advanced the reasoning capabilities of large language models. However, standard Group Relative Policy Optimization (GRPO) typically assigns a uniform, sequence-level advantage to all tokens, thereby overlooking the intrinsic information heterogeneity along reasoning chains. We show that this coarse-grained credit assignment leads to premature entropy collapse and encourages the model to generate redundant, low-quality reasoning paths. Through systematic empirical analysis, we identify Critical Decision Pivots (CDPs): transient high-entropy states where the policy's trajectory is most sensitive to perturbations. These pivots represent the "forks in the road" where effective multi-path exploration is most crucial yet often suppressed by uniform advantage signals. Building on these insights, we propose Entropy-Regulated Policy Optimization (ERPO), which transitions the optimization focus from coarse sequences to fine-grained token dynamics. ERPO introduces three synergistic components: (i) Entropy-aware Gating, which adaptively amplifies exploration at CDPs to facilitate diverse path discovery; (ii) Bucket-based Implicit Normalization, which mitigates difficulty bias by aligning token progress windows; and (iii) Result-anchored Advantage Synthesis, which re-weights token-level signals via outcome-driven anchors. Extensive experiments on competitive mathematical benchmarks (e.g., MATH, AIME) demonstrate that ERPO significantly outperforms GRPO. Notably, ERPO not only boosts reasoning accuracy but also yields significantly more concise and robust derivation paths, establishing a new efficiency-accuracy frontier for large reasoning models.

Jiahao Lou, Shufeng Gong, Quan Yu et al.

On-disk graph-based indexes are favored for billion-scale Approximate Nearest Neighbor Search (ANNS) due to their high performance and cost-efficiency. However, existing systems typically rely on a coupled storage architecture that co-locates vectors and graph topology, which introduces substantial redundant I/O during index updates, thereby degrading usability in dynamic workloads. In this paper, we propose a decoupled storage architecture that physically separates heavy vectors from the lightweight graph topology. This design substantially improves update performance by reducing redundant I/O during updates. However, it introduces I/O amplification during ANNS, leading to degraded query efficiency.To improve query performance within the update-friendly architecture, we propose two techniques co-designed with the decoupled storage. We develop a similarity-aware dynamic layout that optimizes data placement online so that redundantly fetched data can be reused in subsequent search steps, effectively turning read amplification into useful prefetching. In addition, we propose a two-stage query mechanism enhanced by hierarchical PQ, which uses hierarchical PQ to rapidly and accurately identify promising candidates and performs exact refinement on raw vectors for only a small number of candidates. This design significantly reduces both the I/O and computational cost of the refinement stage. Overall, DGAI achieves resource-efficient updates and low-latency queries simultaneously. Experimental results demonstrate that \oursys improves update speed by 8.17x for insertions and 8.16x for deletions, while reducing peak query latency under mixed workloads by 67\% compared to state-of-the-art baselines.

Song Yu, Xiaofei Xu, Fangfei Xu et al.

Although large language models perform well in understanding and responding to user intent, their performance in specialized domains such as Traditional Chinese Medicine (TCM) remains limited due to lack of expertise. In addition, high-quality data related to TCM is scarce and difficult to obtain, making large language models ineffective in handling TCM tasks. In this work, we propose a framework to improve the performance of large language models for TCM tasks using only a small amount of data. First, we use medical case data for supervised fine-tuning of the large model, making it initially capable of performing TCM tasks. Subsequently, we further optimize the model's performance using reinforcement learning from AI feedback (RLAIF) to align it with the preference data. The ablation study also demonstrated the performance gain is attributed to both supervised fine-tuning and the direct policy optimization. The experimental results show that the model trained with a small amount of data achieves a significant performance improvement on a representative TCM task.

Kehui Tan, Tianqi Pang, Chenyou Fan et al.

This perspective paper proposes a series of interactive scenarios that utilize Artificial Intelligence (AI) to enhance classroom teaching, such as dialogue auto-completion, knowledge and style transfer, and assessment of AI-generated content. By leveraging recent developments in Large Language Models (LLMs), we explore the potential of AI to augment and enrich teacher-student dialogues and improve the quality of teaching. Our goal is to produce innovative and meaningful conversations between teachers and students, create standards for evaluation, and improve the efficacy of AI-for-Education initiatives. In Section 3, we discuss the challenges of utilizing existing LLMs to effectively complete the educated tasks and present a unified framework for addressing diverse education dataset, processing lengthy conversations, and condensing information to better accomplish more downstream tasks. In Section 4, we summarize the pivoting tasks including Teacher-Student Dialogue Auto-Completion, Expert Teaching Knowledge and Style Transfer, and Assessment of AI-Generated Content (AIGC), providing a clear path for future research. In Section 5, we also explore the use of external and adjustable LLMs to improve the generated content through human-in-the-loop supervision and reinforcement learning. Ultimately, this paper seeks to highlight the potential for AI to aid the field of education and promote its further exploration.

Zhifang Liao, Haihui Pan, Qi Zeng et al.

As an important part of the power system, power load forecasting directly affects the national economy. The data shows that improving the load forecasting accuracy by 0.01% can save millions of dollars for the power industry. Therefore, improving the accuracy of power load forecasting has always been the pursuing goals for a power system. Based on this goal, this paper proposes a novel connection, the dense average connection, in which the outputs of all preceding layers are averaged as the input of the next layer in a feed-forward fashion. Based on dense average connection , we construct the dense average network for power load forecasting. The predictions of the proposed model for two public datasets are better than those of existing methods. On this basis, we use the ensemble method to further improve the accuracy of the model. To verify the reliability of the model predictions, the robustness is analyzed and verified by adding input disturbances. The experimental results show that the proposed model is effective and robust for power load forecasting.