arXiv subjects

Luozhijie Jin

Publications and source records attributed to Luozhijie Jin.

9 recordsLinked to original sources

arXiv2026

ContextWeave: A Real-World Workflow Benchmark

Memory is essential as language agents move from isolated tasks to long-horizon, stateful workflows, yet existing evaluations often reduce it to retrieval or question answering. We introduce ContextWeave, a longitudinal benchmark that evaluates whether recalled experience improves downstream agent performance in realistic office-work streams. ContextWeave reconstructs privacy-preserved, multi-month workflows of 14 participants into 1,005 executable tasks, including 568 core evaluation tasks, with instructions, containerized environments, trajectories, and task-specific rubrics. It measures workspace quality and alignment with participant-specific preferences, complemented by diagnostics of relevance, continuity, solvability, and robustness to misleading recall. Across six memory components under a fixed model, the strongest configuration raises Workspace Score from 68.08 to 78.20 and Preference Score from 41.50 to 70.60. With a fixed memory component, recall improves both outcomes for all five tested base models, although gains vary substantially. Our analysis shows that actionable, experience-rich memory supports workflow continuation and reduces redundant exploration more effectively than compact summaries, while it can also be more susceptible to misleading recall. These findings motivate memory systems that optimize not only retrieval relevance but also reliable use during execution.

cs.AI↗

arXiv2025

From Rational Answers to Emotional Resonance: The Role of Controllable Emotion Generation in Language Models

Purpose: Emotion is a fundamental component of human communication, shaping understanding, trust, and engagement across domains such as education, healthcare, and mental health. While large language models (LLMs) exhibit strong reasoning and knowledge generation capabilities, they still struggle to express emotions in a consistent, controllable, and contextually appropriate manner. This limitation restricts their potential for authentic human-AI interaction. Methods: We propose a controllable emotion generation framework based on Emotion Vectors (EVs) - latent representations derived from internal activation shifts between neutral and emotion-conditioned responses. By injecting these vectors into the hidden states of pretrained LLMs during inference, our method enables fine-grained, continuous modulation of emotional tone without any additional training or architectural modification. We further provide theoretical analysis proving that EV steering enhances emotional expressivity while maintaining semantic fidelity and linguistic fluency. Results: Extensive experiments across multiple LLM families show that the proposed approach achieves consistent emotional alignment, stable topic adherence, and controllable affect intensity. Compared with existing prompt-based and fine-tuning-based baselines, our method demonstrates superior flexibility and generalizability. Conclusion: Emotion Vector (EV) steering provides an efficient and interpretable means of bridging rational reasoning and affective understanding in large language models, offering a promising direction for building emotionally resonant AI systems capable of more natural human-machine interaction.

cs.CL↗

arXiv2025

MOSS-Speech: Towards True Speech-to-Speech Models Without Text Guidance

Spoken dialogue systems often rely on cascaded pipelines that transcribe, process, and resynthesize speech. While effective, this design discards paralinguistic cues and limits expressivity. Recent end-to-end methods reduce latency and better preserve these cues, yet still rely on text intermediates, creating a fundamental bottleneck. We present MOSS-Speech, a true speech-to-speech large language model that directly understands and generates speech without relying on text guidance. Our approach combines a modality-based layer-splitting architecture with a frozen pre-training strategy, preserving the reasoning and knowledge of pretrained text LLMs while adding native speech capabilities. Experiments show that our model achieves state-of-the-art results in spoken question answering and delivers comparable speech-to-speech performance relative to existing text-guided systems, while still maintaining competitive text performance. By narrowing the gap between text-guided and direct speech generation, our work establishes a new paradigm for expressive and efficient end-to-end speech interaction.

cs.CL↗

arXiv2025

CodecBench: A Comprehensive Benchmark for Acoustic and Semantic Evaluation

With the rise of multimodal large language models (LLMs), audio codec plays an increasingly vital role in encoding audio into discrete tokens, enabling integration of audio into text-based LLMs. Current audio codec captures two types of information: acoustic and semantic. As audio codec is applied to diverse scenarios in speech language model , it needs to model increasingly complex information and adapt to varied contexts, such as scenarios with multiple speakers, background noise, or richer paralinguistic information. However, existing codec's own evaluation has been limited by simplistic metrics and scenarios, and existing benchmarks for audio codec are not designed for complex application scenarios, which limits the assessment performance on complex datasets for acoustic and semantic capabilities. We introduce CodecBench, a comprehensive evaluation dataset to assess audio codec performance from both acoustic and semantic perspectives across four data domains. Through this benchmark, we aim to identify current limitations, highlight future research directions, and foster advances in the development of audio codec. The codes are available at https://github.com/RayYuki/CodecBench.

eess.AS↗

arXiv2025

Inference-Time Alignment Control for Diffusion Models with Reinforcement Learning Guidance

Denoising-based generative models, particularly diffusion and flow matching algorithms, have achieved remarkable success. However, aligning their output distributions with complex downstream objectives, such as human preferences, compositional accuracy, or data compressibility, remains challenging. While reinforcement learning (RL) fine-tuning methods, inspired by advances in RL from human feedback (RLHF) for large language models, have been adapted to these generative frameworks, current RL approaches are suboptimal for diffusion models and offer limited flexibility in controlling alignment strength after fine-tuning. In this work, we reinterpret RL fine-tuning for diffusion models through the lens of stochastic differential equations and implicit reward conditioning. We introduce Reinforcement Learning Guidance (RLG), an inference-time method that adapts Classifier-Free Guidance (CFG) by combining the outputs of the base and RL fine-tuned models via a geometric average. Our theoretical analysis shows that RLG's guidance scale is mathematically equivalent to adjusting the KL-regularization coefficient in standard RL objectives, enabling dynamic control over the alignment-quality trade-off without further training. Extensive experiments demonstrate that RLG consistently improves the performance of RL fine-tuned models across various architectures, RL algorithms, and downstream tasks, including human preferences, compositional control, compressibility, and text rendering. Furthermore, RLG supports both interpolation and extrapolation, thereby offering unprecedented flexibility in controlling generative alignment. Our approach provides a practical and theoretically sound solution for enhancing and controlling diffusion model alignment at inference. The source code for RLG is publicly available at the Github: https://github.com/jinluo12345/Reinforcement-learning-guidance.

cs.LG↗

arXiv2025

XY-Tokenizer: Mitigating the Semantic-Acoustic Conflict in Low-Bitrate Speech Codecs

Speech codecs serve as bridges between speech signals and large language models. An ideal codec for speech language models should not only preserve acoustic information but also capture rich semantic information. However, existing speech codecs struggle to balance high-quality audio reconstruction with ease of modeling by language models. In this study, we analyze the limitations of previous codecs in balancing semantic richness and acoustic fidelity. We propose XY-Tokenizer, a novel codec that mitigates the conflict between semantic and acoustic capabilities through multi-stage, multi-task learning. Experimental results demonstrate that XY-Tokenizer achieves performance in both semantic and acoustic tasks comparable to that of state-of-the-art codecs operating at similar bitrates, even though those existing codecs typically excel in only one aspect. Specifically, XY-Tokenizer achieves strong text alignment, surpassing distillation-based semantic modeling methods such as SpeechTokenizer and Mimi, while maintaining a speaker similarity score of 0.83 between reconstructed and original audio. The reconstruction performance of XY-Tokenizer is comparable to that of BigCodec, the current state-of-the-art among acoustic-only codecs, which achieves a speaker similarity score of 0.84 at a similar bitrate. Code and models are available at https://github.com/gyt1145028706/XY-Tokenizer.

cs.SD↗

arXiv2024

VQCrystal: Leveraging Vector Quantization for Discovery of Stable Crystal Structures

Discovering functional crystalline materials through computational methods remains a formidable challenge in materials science. Here, we introduce VQCrystal, an innovative deep learning framework that leverages discrete latent representations to overcome key limitations in current approaches to crystal generation and inverse design. VQCrystal employs a hierarchical VQ-VAE architecture to encode global and atom-level crystal features, coupled with a machine learning-based inter-atomic potential(IAP) model and a genetic algorithm to realize property-targeted inverse design. Benchmark evaluations on diverse datasets demonstrate VQCrystal's advanced capabilities in representation learning and novel crystal discovery. Notably, VQCrystal achieves state-of-the-art performance with 91.93\% force validity and a Fréchet Distance of 0.152 on MP-20, indicating both strong validity and high diversity in the sampling process. To demonstrate real-world applicability, we apply VQCrystal for both 3D and 2D material design. For 3D materials, the density-functional theory validation confirmed that 63.04\% of bandgaps and 99\% of formation energies of the 56 filtered materials matched the target range. Moreover, 437 generated materials were validated as existing entries in the full database outside the training set. For the discovery of 2D materials, 73.91\% of 23 filtered structures exhibited high stability with formation energies below -1 eV/atom. Our results highlight VQCrystal's potential to accelerate the discovery of novel materials with tailored properties.

cond-mat.mtrl-sci↗

arXiv2024

CTGNN: Crystal Transformer Graph Neural Network for Crystal Material Property Prediction

The combination of deep learning algorithm and materials science has made significant progress in predicting novel materials and understanding various behaviours of materials. Here, we introduced a new model called as the Crystal Transformer Graph Neural Network (CTGNN), which combines the advantages of Transformer model and graph neural networks to address the complexity of structure-properties relation of material data. Compared to the state-of-the-art models, CTGNN incorporates the graph network structure for capturing local atomic interactions and the dual-Transformer structures to model intra-crystal and inter-atomic relationships comprehensively. The benchmark carried on by the proposed CTGNN indicates that CTGNN significantly outperforms existing models like CGCNN and MEGNET in the prediction of formation energy and bandgap properties. Our work highlights the potential of CTGNN to enhance the performance of properties prediction and accelerates the discovery of new materials, particularly for perovskite materials.

cond-mat.mtrl-sci↗

arXiv2024

Crystal Transformer Based Universal Atomic Embedding for Accurate and Transferable Prediction of Materials Properties

In this work, we propose a novel approach to generate universal atomic embeddings, significantly enhancing the representational and accuracy aspects of atomic embeddings, which ultimately improves the accuracy of property prediction. Moreover, we demonstrate the excellent transferability of universal atomic embeddings across different databases and various property tasks. Our approach centers on developing the CrystalTransformer model. Unlike traditional methods, this model does not possess a fundamental graph network architecture but utilizes the Transformer architecture to extract latent atomic features. This allows the CrystalTransformer to mitigate the inherent topological information bias of graph neural networks while maximally preserving the atomic chemical information, making it more accurate in encoding complex atomic features and thereby offering a deeper understanding of the atoms in materials. In our research, we highlight the advantages of CrystalTransformer in generating universal atomic embeddings through comparisons with current mainstream graph neural network models. Furthermore, we validate the effectiveness of universal atomic embeddings in enhancing the accuracy of model predictions for properties and demonstrate their transferability across different databases and property tasks through various experiments. As another key aspect of our study, we discover the strong physical interpretability implied in universal atomic embeddings through clustering and correlation analysis, indicating the immense potential of our universal atomic embeddings as atomic fingerprints.

cond-mat.mtrl-sci↗