arXiv subjects

Jialong Xue

Publications and source records attributed to Jialong Xue.

4 recordsLinked to original sources

arXiv2026

StepAudio 3 Realtime Technical Report

Realtime spoken interaction demands deep reasoning, prompt responses, and fluid turn-taking. We present StepAudio 3 Realtime, an audio-language foundation model organized around a continuous listen-converse-think-act loop. Deep Perception captures rich acoustic cues to interpret user intent, while Seamless Duplex models synchronized audio streams to handle pauses, backchannels, and interruptions naturally. Crucially, we resolve the tension between deep deliberation and latency via Think-While-Speaking, executing private reasoning in parallel with spoken delivery. In reasoning mode, StepAudio 3 reaches a 73.0 macro average on StepAudioChat. With Think-While-Speaking, it achieves dialogue and reasoning performance comparable to dedicated reasoning models while speaking in real time. Furthermore, an integrated Voice Agent handles asynchronous tool execution without disrupting the dialogue flow. StepAudio 3 Realtime achieves top-tier performance across key dimensions: an exceptional 90.6 on the MMSU benchmark, 98.9 Overall on the Artificial Analysis Full-Duplex Bench, and a 56.0% macro task-success rate on $τ$-Voice.

cs.SD↗

arXiv2026

Unified Spatiotemporal Token Compression for Video-LLMs at Ultra-Low Retention

Video large language models (Video-LLMs) face high computational costs due to large volumes of visual tokens. Existing token compression methods typically adopt a two-stage spatiotemporal compression strategy, relying on stage-specific metrics and an implicit assumption of spatiotemporal separability. Under extremely low retention ratios, however, such approaches often result in unbalanced allocation and loss of visual evidence essential for question answering. We reformulate token compression as a spatiotemporal allocation task within a global token retention pool. We propose a unified selection mechanism that integrates attention weights and semantic similarity to globally select tokens with high contribution and low redundancy. Unselected tokens are merged via clustering and refilled, preserving information integrity. Inside the LLM, we further introduce text-aware merging to perform secondary compression based on query relevance. Without requiring retraining, our method serves as a plug-and-play module compatible with existing Video-LLMs. Experiments show that retaining only about 2% of visual tokens preserves 90.1% of baseline performance across multiple benchmarks, while reducing FLOPs to roughly 2.6%. These benefits generalize across diverse backbones, decreasing end-to-end inference latency and memory consumption. Our unified spatiotemporal token compression strategy establishes the state-of-the-art in video understanding under ultra-low token retention.

cs.CV↗

arXiv2025

High-Precision Transformer-Based Visual Servoing for Humanoid Robots in Aligning Tiny Objects

High-precision tiny object alignment remains a common and critical challenge for humanoid robots in real-world. To address this problem, this paper proposes a vision-based framework for precisely estimating and controlling the relative position between a handheld tool and a target object for humanoid robots, e.g., a screwdriver tip and a screw head slot. By fusing images from the head and torso cameras on a robot with its head joint angles, the proposed Transformer-based visual servoing method can correct the handheld tool's positional errors effectively, especially at a close distance. Experiments on M4-M8 screws demonstrate an average convergence error of 0.8-1.3 mm and a success rate of 93\%-100\%. Through comparative analysis, the results validate that this capability of high-precision tiny object alignment is enabled by the Distance Estimation Transformer architecture and the Multi-Perception-Head mechanism proposed in this paper.

cs.CV↗

arXiv2024

Reflected Search Poisoning for Illicit Promotion

As an emerging black hat search engine optimization (SEO) technique, reflected search poisoning (RSP) allows a miscreant to free-ride the reputation of high-ranking websites, poisoning search engines with illicit promotion texts (IPTs) in an efficient and stealthy manner, while avoiding the burden of continuous website compromise as required by traditional promotion infections. However, little is known about the security implications of RSP, e.g., what illicit promotion campaigns are being distributed by RSP, and to what extent regular search users can be exposed to illicit promotion texts distributed by RSP. In this study, we conduct the first security study on RSP-based illicit promotion, which is made possible through an end-to-end methodology for capturing, analyzing, and infiltrating IPTs. As a result, IPTs distributed via RSP are found to be large-scale, continuously growing, and diverse in both illicit categories and natural languages. Particularly, we have identified over 11 million distinct IPTs belonging to 14 different illicit categories, with typical examples including drug trading, data theft, counterfeit goods, and hacking services. Also, the underlying RSP cases have abused tens of thousands of high-ranking websites, as well as extensively poisoning all four popular search engines we studied, especially Google Search and Bing. Furthermore, it is observed that benign search users are being exposed to IPTs at a concerning extent. To facilitate interaction with potential customers (victim search users), miscreants tend to embed various types of contacts in IPTs, especially instant messaging accounts. Further infiltration of these IPT contacts reveals that the underlying illicit campaigns are operated on a large scale. All these findings highlight the negative security implications of IPTs and RSPs, and thus call for more efforts to mitigate RSP-driven illicit promotion.

cs.CR↗