arXiv subjects

Guoxin Ma

Publications and source records attributed to Guoxin Ma.

7 recordsLinked to original sources

arXiv2026

Thinking as Compression: Your Reasoning Model is Secretly a Context Compressor

Context compression aims to shorten long context inputs with minimal information loss for LLM inference acceleration. While existing methods have shown promise, they typically rely on complex compression modules or compression-specific training, leaving the intrinsic capabilities of LLMs underexplored. In contrast, this work reveals that a thinking model itself can naturally compress long contexts by organizing task-relevant information. We thus derive Thinking as Compression (TaC), a new compression paradigm that treats thinking itself as compressed context. Without relying on specific dedicated compressor, TaC directly prompts the thinking model to generate thinking traces as the shortened context, already outperforming most representative compression methods. Further, given that raw thinking output may struggle with budget control and shortcut behaviors, we introduce Thinking as Compression Constrained (TaC-C), leveraging a simple reward-driven optimization framework to elicit intrinsic thinking as compact and controllable compressed context. Experiments across four long-context QA benchmarks demonstrate that TaC-C consistently outperforms existing baselines. At 4x and 8x compression ratios, it surpasses the strongest competitor by 17.4% and 23.4% in average F1, and by 15.7% and 21.7% in average Exact Match Score (EM), respectively.

cs.AI↗

arXiv2026

Unleash the Potential of Long Semantic IDs for Generative Recommendation

Semantic ID-based generative recommenders face a granularity-efficiency dilemma between efficient recommendation with short IDs and expressive item modeling with long IDs. To break this dilemma, we propose ACERec, a framework that preserves the semantic richness of long IDs while keeping the recommendation process efficient. Concretely, ACERec employs an Attentive Token Merger to compress long semantic IDs into compact yet faithful latent tokens. To better capture user intent from the compressed semantics, we further introduce a dedicated Intent Token, optimized by a dual-granularity objective that combines token-level generation with item-level intent-semantic alignment. Extensive experiments on nine real-world benchmarks show that ACERec consistently outperforms state-of-the-art methods, yielding average relative improvements of 12.92% in NDCG@10 and 7.49% in Recall@10 over the strongest baselines.

cs.IR↗

arXiv2026

When Less is More: The LLM Scaling Paradox in Context Compression

Scaling up model parameters has long been a prevalent training paradigm driven by the assumption that larger models yield superior generation capabilities. However, under lossy context compression in a compressor--decoder setup, we find a \textbf{\textit{Size-Fidelity Paradox}}: increasing compressor size can lessen the faithfulness of reconstructed contexts though reconstruction error decreases. Across 27 compressor setups spanning model families, scales, and compression rates, we coin this paradox arising from two dominant factors: 1) \textit{knowledge overwriting}: larger models increasingly replace source facts with their own prior beliefs, \textit{e.g.}, ``the white strawberry`` $\to$ ``the red strawberry``; and 2) \textit{semantic drift}: larger models tend to paraphrase or restructure content instead of reproducing it verbatim, \textit{e.g.}, ``Alice hit Bob`` $\to$ ``Bob hit Alice``. Interestingly, this paradox persists across varied settings, with mid-sized compressors often outperforming larger ones in faithful recovery. By analyzing the compressed memory via embedding geometry and reconstruction determinacy, we further reveal that compressors tend to organize memory across broader semantic subspaces, yielding more ambiguous representations prone to overwriting, drift, and weakened recovery. These findings complement existing evaluations of context compression and expose a breakdown of scaling laws when the objective shifts from plausible generation to faithful preservation.

cs.LG↗

arXiv2025

DEER: Disentangled Mixture of Experts with Instance-Adaptive Routing for Generalizable Machine-Generated Text Detection

Detecting machine-generated text has become a critical challenge amid the rapid advancement of LLMs, yet existing detectors degrade severely under domain shift. Through systematic pilot studies, we trace this vulnerability to two fundamental flaws in current generalization strategies, namely the incomplete preservation of domain-specific knowledge during multi-domain training and the misalignment between knowledge retrieval and the detection objective at inference. To address these gaps, we propose DEER, a Disentangled mixturE-of-ExpeRts framework that explicitly decouples domain-local and domain-invariant knowledge into specialized expert modules. Instead of static domain matching, DEER employs a reinforcement learning-driven router that selects expert pathways based on instance-level detection rewards. This task-aligned, domain-agnostic mechanism ensures robust adaptation to unseen distributions by prioritizing detection utility over stylistic resemblance. Extensive experiments demonstrate that DEER consistently outperforms state-of-the-art detectors, achieving average F1 improvements of 1.28% and 2.92%, and accuracy gains of 1.35% and 2.26% on in-domain and out-of-domain datasets, offering reliable generalization for open-world deployment.

cs.CL↗

arXiv2025

Can Reasoning Path still be Effective as Input? Bridging Post-Reasoning to Chain-of-Thought Compression

Recent developments have enabled advanced reasoning in Large Language Models (LLMs) via long Chain-of-Thought (CoT), trading efficiency during inference for performance. Existing works focus on compressing generated CoT in reasoning, which impairs the necessary information for deriving the correct answer. In this work, we propose post-reasoning, a reasoning paradigm that takes CoT as a part of context to simplify the reasoning task for LLMs. We find that post-reasoning significantly reduces the generation length of LLMs, but its effectiveness hinges on the efficiency and the reliability of the contextual CoT generation. Therefore, we propose Upfront CoT (UCoT), an efficient post-reasoning framework for CoT compression. UCoT trains a lightweight model (compressor) to provide contextual CoT in form of soft tokens and trains the LLM (executor) to leverage this contextual CoT for producing the final answer. Extensive experiments show that UCoT maintains the powerful reasoning ability of executor while significantly reducing the length of CoT. It is worth mentioning that when applying UCoT to the Qwen2.5-7B-Instruct model, the usage of tokens on GSM8K dataset is reduced by 50%, while the performance is 3.08% higher than that of the state-of-the-art (SOTA) method.

cs.CL↗

arXiv2025

MGT-Prism: Enhancing Domain Generalization for Machine-Generated Text Detection via Spectral Alignment

Large Language Models have shown growing ability to generate fluent and coherent texts that are highly similar to the writing style of humans. Current detectors for Machine-Generated Text (MGT) perform well when they are trained and tested in the same domain but generalize poorly to unseen domains, due to domain shift between data from different sources. In this work, we propose MGT-Prism, an MGT detection method from the perspective of the frequency domain for better domain generalization. Our key insight stems from analyzing text representations in the frequency domain, where we observe consistent spectral patterns across diverse domains, while significant discrepancies in magnitude emerge between MGT and human-written texts (HWTs). The observation initiates the design of a low frequency domain filtering module for filtering out the document-level features that are sensitive to domain shift, and a dynamic spectrum alignment strategy to extract the task-specific and domain-invariant features for improving the detector's performance in domain generalization. Extensive experiments demonstrate that MGT-Prism outperforms state-of-the-art baselines by an average of 0.90% in accuracy and 0.92% in F1 score on 11 test datasets across three domain-generalization scenarios.

cs.CL↗

arXiv2024

Attention-based UNet enabled Lightweight Image Semantic Communication System over Internet of Things

This paper studies the problem of the lightweight image semantic communication system that is deployed on Internet of Things (IoT) devices. In the considered system model, devices must use semantic communication techniques to support user behavior recognition in ultimate video service with high data transmission efficiency. However, it is computationally expensive for IoT devices to deploy semantic codecs due to the complex calculation processes of deep learning (DL) based codec training and inference. To make it affordable for IoT devices to deploy semantic communication systems, we propose an attention-based UNet enabled lightweight image semantic communication (LSSC) system, which achieves low computational complexity and small model size. In particular, we first let the LSSC system train the codec at the edge server to reduce the training computation load on IoT devices. Then, we introduce the convolutional block attention module (CBAM) to extract the image semantic features and decrease the number of downsampling layers thus reducing the floating-point operations (FLOPs). Finally, we experimentally adjust the structure of the codec and find out the optimal number of downsampling layers. Simulation results show that the proposed LSSC system can reduce the semantic codec FLOPs by 14%, and reduce the model size by 55%, with a sacrifice of 3% accuracy, compared to the baseline. Moreover, the proposed scheme can achieve a higher transmission accuracy than the traditional communication scheme in the low channel signal-to-noise (SNR) region.

cs.NE↗