arXiv subjects

Yichen Wan

Publications and source records attributed to Yichen Wan.

5 recordsLinked to original sources

arXiv2026

DuCCAE: A Hybrid Engine for Immersive Conversation via Collaboration, Augmentation, and Evolution

Immersive conversational systems in production face a persistent trade-off between responsiveness and long-horizon task capability. Real-time interaction is achievable for lightweight turns, but requests involving planning and tool invocation (e.g., search and media generation) produce heavy-tail execution latency that degrades turn-taking, persona consistency, and user trust. To address this challenge, we propose DuCCAE (Conversation while Collaboration with Augmentation and Evolution), a hybrid engine for immersive conversation deployed within Baidu Search, serving millions of users. DuCCAE decouples real-time response generation from asynchronous agentic execution and synchronizes them via a shared state that maintains session context and execution traces, enabling asynchronous results to be integrated back into the ongoing dialogue. The system orchestrates five subsystems-Info, Conversation, Collaboration, Augmentation, and Evolution-to support multi-agent collaboration and continuous improvement. We evaluate DuCCAE through a comprehensive framework that combines offline benchmarking on the Du-Interact dataset and large-scale production evaluation within Baidu Search. Experimental results demonstrate that DuCCAE outperforms strong baselines in agentic execution reliability and dialogue quality while reducing latency to fit strict real-time budgets. Crucially, deployment metrics since June 2025 confirm substantial real-world effectiveness, evidenced by a tripling of Day-7 user retention to 34.2% and a surge in the complex task completion rate to 65.2%. Our hybrid architecture successfully preserves conversational continuity while enabling reliable agentic execution, offering practical guidelines for deploying scalable agentic systems in industrial settings.

cs.CL↗

arXiv2026

Stay in Character, Stay Safe: Dual-Cycle Adversarial Self-Evolution for Safety Role-Playing Agents

LLM-based role-playing has rapidly improved in fidelity, yet stronger adherence to persona constraints commonly increases vulnerability to jailbreak attacks, especially for risky or negative personas. Most prior work mitigates this issue with training-time solutions (e.g., data curation or alignment-oriented regularization). However, these approaches are costly to maintain as personas and attack strategies evolve, can degrade in-character behavior, and are typically infeasible for frontier closed-weight LLMs. We propose a training-free Dual-Cycle Adversarial Self-Evolution framework with two coupled cycles. A Persona-Targeted Attacker Cycle synthesizes progressively stronger jailbreak prompts, while a Role-Playing Defender Cycle distills observed failures into a hierarchical knowledge base of (i) global safety rules, (ii) persona-grounded constraints, and (iii) safe in-character exemplars. At inference time, the Defender retrieves and composes structured knowledge from this hierarchy to guide generation, producing responses that remain faithful to the target persona while satisfying safety constraints. Extensive experiments across multiple proprietary LLMs show consistent gains over strong baselines on both role fidelity and jailbreak resistance, and robust generalization to unseen personas and attack prompts.

cs.AI↗

arXiv2024

Interfacial fatigue fracture of pressure sensitive adhesives

Pressure sensitive adhesives (PSAs) are viscoelastic polymers that can form fast and robust adhesion with various adherends under fingertip pressure. The rapidly expanding application domain of PSAs, such as healthcare, wearable electronics, and flexible displays, requires PSAs to sustain prolonged loads throughout their lifetime, calling for fundamental studies on their fatigue behaviors. However, fatigue of PSAs has remained poorly investigated. Here we study interfacial fatigue fracture of PSAs, focusing on the cyclic interfacial crack propagation due to the gradual rupture of noncovalent bonds between a PSA and an adherend. We fabricate a model PSA made of a hysteresis-free poly(butyl acrylate) bulk elastomer dip-coated with a viscoelastic poly(butyl acrylate-co-isobornyl acrylate) sticky surface, both crosslinked by poly(ethylene glycol) diacrylate. We adhere the fabricated PSA to a polyester strip to form a bilayer. The bilayer is covered by another polyester film as an inextensible backing layer. Using cyclic and monotonic peeling tests, we characterize the interfacial fatigue and fracture behaviors of the bilayer. From the experimental data, we obtain the interfacial fatigue threshold (4.6 J/m2) under cyclic peeling, the slow crack threshold (33.9 J/m2) under monotonic peeling, and the adhesion toughness (~ 400 J/m2) at a finite peeling speed. We develop a modified Lake-Thomas model to describe the interfacial fatigue threshold due to noncovalent bond breaking. The theoretical prediction (2.6 J/m2) agrees well with the experimental measurement (4.6 J/m2). Finally, we discuss possible additional dissipation mechanisms involved in the larger slow crack threshold and much larger adhesion toughness. It is hoped that this study will provide new fundamental knowledge for fracture mechanics of PSAs, as well as guidance for future tough and durable PSAs.

cond-mat.soft↗

arXiv2024

Strong and Highly Switchable Soft Sticky Adhesives

Many biological systems can form strong adhesion to various materials with complex shapes. The adhesion is further switchable between strongly adhering and completely non-adhering in a simple and fast manner. By contrast, no engineering system has yet achieved the same robust adherence and switching. This limitation severely hinders the advancement of several emerging technologies including biomimetic robots, assembly-based manufacturing, precision medicine, wearable and implantable devices, as well as on-demand material dismantling and recycling for sustainability. Here we present a design approach for strong and highly switchable adhesion by synergizing the surface stickiness, bulk energy dissipation, and stimuli-responsive polymer chains in a thermo-switchable soft sticky adhesive. The adhesive has a high adhesion strength of about 80 kPa with diverse materials at room temperature. The adhesion is highly switchable to near-vanishing (about 0.6 kPa) at an elevated temperature due to the thermo-responsive surface polymer chain retraction. This adhesion switching is reversible and repeatable for many cycles, enabling selective pick-and-release of objects with various materials, shapes, sizes, and weights. The switching time is around 10 s with an adhesive layer of 1 mm, governed by thermal conduction through the adhesive, faster than or comparable to most state-of-the-art methods. The adhesive is self-healing, and can be recycled, dried, stored, reswollen, and reused with nearly intact adhesion and switching properties. The synergistic design combining strong adhesion and stimuli-responsive switching can be potentially extended to various polymer systems, and further enhanced by optimized surface architectures.

cond-mat.soft↗

arXiv2023

Data and Model Poisoning Backdoor Attacks on Wireless Federated Learning, and the Defense Mechanisms: A Comprehensive Survey

Due to the greatly improved capabilities of devices, massive data, and increasing concern about data privacy, Federated Learning (FL) has been increasingly considered for applications to wireless communication networks (WCNs). Wireless FL (WFL) is a distributed method of training a global deep learning model in which a large number of participants each train a local model on their training datasets and then upload the local model updates to a central server. However, in general, non-independent and identically distributed (non-IID) data of WCNs raises concerns about robustness, as a malicious participant could potentially inject a "backdoor" into the global model by uploading poisoned data or models over WCN. This could cause the model to misclassify malicious inputs as a specific target class while behaving normally with benign inputs. This survey provides a comprehensive review of the latest backdoor attacks and defense mechanisms. It classifies them according to their targets (data poisoning or model poisoning), the attack phase (local data collection, training, or aggregation), and defense stage (local training, before aggregation, during aggregation, or after aggregation). The strengths and limitations of existing attack strategies and defense mechanisms are analyzed in detail. Comparisons of existing attack methods and defense designs are carried out, pointing to noteworthy findings, open challenges, and potential future research directions related to security and privacy of WFL.

cs.CR↗