The decline of proficient practitioners poses an existential threat to numerous ancient writing systems, risking the loss of the cultural heritage they encode. Among these systems is Sawndip, a logographic script whose computational preservation faces severe challenges due to the scarcity of standardized corpora. In this paper, we present the first systematic framework for low-resource Sawndip-Chinese translation and bilingual conversational modeling. We introduce a newly curated dataset comprising sentence-level Sawndip-Chinese parallel data and character-level lexical definitions. Building upon this foundation, we propose a knowledge-enhanced supervised fine-tuning paradigm that incorporates lexical knowledge during training. To address the additional challenges of Chinese-to-Sawndip generation, we further develop a manually curated template-based substructure substitution strategy for synthesizing reverse-direction data. Bilingual conversational data is then constructed through a dimension-conditioned dialogue generation and multi-candidate back-translation pipeline, enabling controllable and diverse dialogue synthesis. Experimental results show that our final unified model effectively supports both Sawndip-Chinese translation and bilingual conversational abilities. Ultimately, this work provides a foundational reference for the digital preservation of Sawndip, offering broader insights into the computational documentation of globally-endangered languages.

Group Relative Policy Optimization (GRPO) significantly enhances the reasoning performance of Large Language Models (LLMs). However, this success heavily relies on expensive external verifiers or human rules. Such dependency not only leads to significant computational costs and training latency, but also yields sparse rewards that hinder optimization efficiency. To address these challenges, we propose Latent-GRPO, a framework that derives intrinsic rewards directly from latent space geometry. Crucially, our empirical analysis reveals a compelling geometric property: terminal token representations of correct reasoning trajectories form dense clusters with high intra-class similarity, whereas incorrect trajectories remain scattered as outliers. In light of this discovery, we introduce the Iterative Robust Centroid Estimation (IRCE) algorithm, which generates dense, continuous rewards by mitigating magnitude fluctuations via spherical projection and estimating a robust “truth centroid” through iterative aggregation. Experimental results on multiple datasets show that our method maintains model performance while achieving a training speedup of over 2x compared to baselines. Furthermore, extensive results demonstrate strong generalization ability and robustness.

Nowadays, developing reliable DeepResearch-style long-form report generation remains challenging, as training and evaluation lack verifiable reward signals. Accordingly, rubric-based evaluation has become a common practice. However, existing approaches either rely on coarse, pre-defined rubrics that lack sufficient granularity or depend on manually constructed query-specific rubrics that are costly and difficult to scale. In this paper, we propose a pipeline to train preference-grounded query-specific rubric generators tailored for DeepResearch report generation. We first construct a dataset of DeepResearch-style queries annotated with human preferences over paired reports, and train rubric generators via reinforcement learning with a hybrid reward combining preference consistency, format validity, and LLM-based rubric evaluation. We evaluate the resulting rubric generators in two stages. First, on a held-out human-preference test set, the learned rubrics discriminate preferred from rejected reports more effectively than generic, prompted, or SFT-trained rubric alternatives. Second, when used as reward signals to train DeepResearch systems, our rubric generators yield substantial performance gains under both a simple single-agent ReAct framework and a complex multi-agent workflow on the DeepResearch Bench. Our code and model are available at https://huggingface.co/fdu-lcz/rubric_generator.

Real-time voice agents face a tension: end-to-end models often lack deep reasoning, while cascaded ASR–LLM–TTS pipelines achieve stronger reasoning but defer LLM inference until after ASR finalizes the transcript, creating an avoidable post–turn-end silent gap. We target this gap under complex spoken requests that admit a relatively determinate answer, and treat it as system-induced rather than communicatively motivated. Existing cascaded streaming strategies attempt to reduce it via mechanical segmentation (e.g., fixed chunks, VAD-based splitting) or speculative generation, but they frequently either break semantic units or waste computation on predictions that must be rolled back. We propose LTS-VoiceAgent, a Listen–Think–Speak framework that explicitly separates when to think from how to reason incrementally. It features a Dynamic Semantic Trigger to detect meaningful prefixes, and a Dual-Role Stream Orchestrator that coordinates a background Thinker (for state maintenance) and a foreground Speaker (for speculative solving). This parallel design enables "thinking while listening" without committing audible content before turn end. We also introduce a Pause-and-Repair benchmark that injects natural disfluencies as a semantic-level streaming stress test. Experiments across VERA, Spoken-MQA, BigBenchAudio, and our benchmark show that LTS-VoiceAgent achieves a stronger accuracy–latency–efficiency trade-off than serial cascaded baselines and existing streaming strategies.

As LLM-based agents are increasingly used in long-term interactions, cumulative memory is critical for enabling personalization and maintaining stylistic consistency. However, most existing systems adopt an “all-or-nothing” approach to memory usage: incorporating all relevant past information can lead to Memory Anchoring, where the agent is trapped by past interactions, while excluding memory entirely results in under-utilization and the loss of important interaction history. We show that an agent’s reliance on memory can be modeled as an explicit and user-controllable dimension. We first introduce a behavioral metric of memory dependence to quantify the influence of past interactions on current outputs. We then propose Steerable Memory Agent, SteeM, a framework that allows users to dynamically regulate memory reliance, ranging from a fresh-start mode that promotes innovation to a high-fidelity mode that closely follows interaction history. Experiments across different scenarios demonstrate that our approach consistently outperforms conventional prompting and rigid memory masking strategies, yielding a more nuanced and effective control for personalized human-agent collaboration.

The primary goal of traditional federated learning is to protect data privacy by enabling distributed edge devices to collaboratively train a shared global model while keeping raw data decentralized at local clients. The rise of large language models (LLMs) has introduced new challenges in distributed systems, as their substantial computational requirements and the need for specialized expertise raise critical concerns about protecting intellectual property (IP). This highlights the need for a federated learning approach that can safeguard both sensitive data and proprietary models. To tackle this challenge, we propose FedQSN, a federated learning approach that leverages random masking to obscure a subnetwork of model parameters and applies quantization to the remaining parameters. Consequently, the server transmits only a privacy-preserving proxy of the global model to clients during each communication round, thus enhancing the model’s confidentiality. Experimental results across various models and tasks demonstrate that our approach not only maintains strong model performance in federated learning settings but also achieves enhanced protection of model parameters compared to baseline methods.

Few-shot knowledge graph completion is to infer the unknown facts (i.e., query head-tail entity pairs) of a given relation with only a few observed reference entity pairs. Its general process is to first encode the implicit relation of an entity pair and then match the relation of a query entity pair with the relations of the reference entity pairs. Most existing methods have thus far encoded an entity pair and matched entity pairs by using the direct neighbors of concerned entities. In this paper, we propose the P-INT model for effective few-shot knowledge graph completion. First, P-INT infers and leverages the paths that can expressively encode the relation of two entities. Second, to capture the fine grained matches, P-INT calculates the interactions of paths instead of mix- ing them for each entity pair. Extensive experimental results demonstrate that P-INT out- performs the state-of-the-art baselines by 11.2– 14.2% in terms of Hits@1. Our codes and datasets are online now.