arXiv subjects

Wei-neng Chen

Publications and source records attributed to Wei-neng Chen.

5 recordsLinked to original sources

arXiv2026

When Does a Language Model Commit? A Finite-Answer Theory of Pre-Verbalization Commitment

Language models often generate reasoning before giving a final answer, but the visible answer does not reveal when the model's answer preference became stable. We study this question through a narrow computable object: \emph{finite-answer preference stabilization}. For a model state and specified answer verbalizers, we project the model's own continuation probabilities onto a finite answer set; in binary tasks this yields an exact log-odds code, $δ(ξ)=S_θ(\mathrm{yes}\midξ)-S_θ(\mathrm{no}\midξ)$. This target defines parser-based answer onset, retrospective stabilization time, and lead without relying on greedy rollouts or learned probes. In controlled delayed-verdict tasks with Qwen3-4B-Instruct, the contextual finite-answer projection stabilizes before the answer is parseable, with 17--31 token mean lead in the main templates and positive, shorter lead in a parser-clean replication. The signal tracks the model's eventual output rather than truth, is linearly recoverable from compact hidden summaries, is partly separable from cursor progress, and transfers as shared information without a single invariant coordinate. Diagnostics separate the measurement from online stopping, verbalizer-free belief, and causal answer control; exact steering shows local sensitivity of $δ$ but not reliable generation control.

cs.AI↗

arXiv2026

Authority Inversion in LLM-Mediated Ubiquitous Systems: When Models Trust Users Over Sensors

Large language models (LLMs) increasingly fuse heterogeneous inputs in ubiquitous systems. Yet, how LLMs implicitly allocate authority when sensor measurements and user claims conflict remains unexamined, raising critical reliability concerns for deployments where physical sensing must retain priority. Unlike explicit traditional fusion, LLMs bury authority allocation within learned representations. We discover this allocation is severely format-dependent: numerical sensor data fails to integrate into answer-relevant model directions, allowing natural-language claims to dominate the final decision, a phenomenon we term \textbf{Authority Inversion}.To diagnose and mitigate this, we develop a geometric framework of context integration, introduce two computable audit metrics, specifically the Context Integration Ratio (CIR) and Authority Alignment Index (AAI), and propose Geometric Authority Calibration (GAC), an inference-time layer-level intervention to suppress misplaced user authority. Evaluating four models (4B to 35B parameters, three architectures) across four datasets totaling 576 conflict instances reveals extreme inversion: on numerical tasks, models exhibit near-zero sensor trust (AAI = -0.805, Cohen's d = -2.14), unaffected by model capacity. Validating our geometric framework, theory-guided causal injection flips 80.2\% of incorrect decisions (vs. <0.4\% for random controls). Practically, GAC improves HAR accuracy from 0 -- 1.6\% to 21.9 -- 27.5\%, outperforming prompting baselines. Ultimately, authority allocation in LLM-mediated systems must be explicitly audited and application-specifically configured rather than left implicit.

cs.AI↗

arXiv2026

The Geometric Price of Discrete Logic: Context-driven Manifold Dynamics of Number Representations

Large language models (LLMs) generalize smoothly across continuous semantic spaces, yet strict logical reasoning demands the formation of discrete decision boundaries. Prevailing theories relying on linear isometric projections fail to resolve this fundamental tension. In this work, we argue that task context operates as a non-isometric dynamical operator that enforces a necessary "topological distortion." By applying Gram-Schmidt decomposition to residual-stream activations , we reveal a dual-modulation mechanism driving this process: a class-agnostic topological preservation that anchors global structure to prevent semantic collapse, and a specific algebraic divergence that directionally tears apart cross-class concepts to forge logical boundaries. We validate this geometric evolution across a gradient of tasks, from simple mapping to complex primality testing. Crucially, targeted specific vector ablation establishes a strict causal binding between this topology and model function: algebraically erasing the divergence component collapses parity classification accuracy from 100% to chance levels (38.57%). Furthermore, we uncover a three-phase layer-wise geometric dynamic and demonstrate that under social pressure prompts, models fail to generate sufficient divergence. This results in a "manifold entanglement" that geometrically explains sycophancy and hallucination. Ultimately, our findings revise the linear-isometric presumption, demonstrating that the emergence of discrete logic in LLMs is purchased at an irreducible cost of topological deformation.

cs.LG↗

arXiv2025

Human-like Social Compliance in Large Language Models: Unifying Sycophancy and Conformity through Signal Competition Dynamics

The increasing integration of Large Language Models (LLMs) into decision-making frameworks has exposed significant vulnerabilities to social compliance, specifically sycophancy and conformity. However, a critical research gap exists regarding the fundamental mechanisms that enable external social cues to systematically override a model's internal parametric knowledge. This study introduces the Signal Competition Mechanism, a unified framework validated by assessing behavioral correlations across 15 LLMs and performing latent-space probing on three representative open-source models. The analysis demonstrates that sycophancy and conformity originate from a convergent geometric manifold, hereafter termed the compliance subspace, which is characterized by high directional similarity in internal representations. Furthermore, the transition to compliance is shown to be a deterministic process governed by a linear boundary, where the Social Emotional Signal effectively suppresses the Information Calibration Signal. Crucially, we identify a "Transparency-Truth Gap," revealing that while internal confidence provides an inertial barrier, it remains permeable and insufficient to guarantee immunity against intense social pressure. By formalizing the Integrated Epistemic Alignment Framework, this research provides a blueprint for transitioning from instructional adherence to robust epistemic integrity.

cs.CY↗

arXiv2021

When Crowdsensing Meets Federated Learning: Privacy-Preserving Mobile Crowdsensing System

Mobile crowdsensing (MCS) is an emerging sensing data collection pattern with scalability, low deployment cost, and distributed characteristics. Traditional MCS systems suffer from privacy concerns and fair reward distribution. Moreover, existing privacy-preserving MCS solutions usually focus on the privacy protection of data collection rather than that of data processing. To tackle faced problems of MCS, in this paper, we integrate federated learning (FL) into MCS and propose a privacy-preserving MCS system, called \textsc{CrowdFL}. Specifically, in order to protect privacy, participants locally process sensing data via federated learning and only upload encrypted training models. Particularly, a privacy-preserving federated averaging algorithm is proposed to average encrypted training models. To reduce computation and communication overhead of restraining dropped participants, discard and retransmission strategies are designed. Besides, a privacy-preserving posted pricing incentive mechanism is designed, which tries to break the dilemma of privacy protection and data evaluation. Theoretical analysis and experimental evaluation on a practical MCS application demonstrate the proposed \textsc{CrowdFL} can effectively protect participants privacy and is feasible and efficient.

cs.CR↗