Searcharxiv⌕ Search

arXiv subjects

Qi Zhang

Publications and source records attributed to Qi Zhang.

At least 271 records · Page 15Linked to original sources
arXiv2026

Analyzing the Effects of Supervised Fine-Tuning on Model Knowledge from Token and Parameter Levels

Large language models (LLMs) acquire substantial world knowledge during pre-training, which is further shaped by post-training techniques such as supervised fine-tuning (SFT). However, the impact of SFT on a model's knowledge remains underexplored, limiting our ability to control knowledge change behavior in fine-tuned models. To address this gap, we evaluate closed-book question answering (CBQA) performance across five LLMs from the LLaMA-2 and LLaMA-3 families. Surprisingly, models fine-tuned on 1,920 samples perform up to 14% worse than those fine-tuned on only 240 samples. Furthermore, varying the level of knowledge mastery in the fine-tuning data leads to performance fluctuations of over 12%. To investigate these effects, we analyze model behavior at both the token and parameter levels. Our analysis reveals that up to 90% of parameter updates during SFT do not contribute to knowledge enhancement. Restoring these updates can improve performance on the CBQA task, depending on the characteristics of the fine-tuning data. These insights offer practical guidance for developing fine-tuning strategies that more effectively strengthen model knowledge.

cs.CL↗
arXiv2026

PTS-SNN: A Prompt-Tuned Temporal Shift Spiking Neural Networks for Efficient Speech Emotion Recognition

Speech Emotion Recognition (SER) is widely deployed in Human-Computer Interaction, yet the high computational cost of conventional models hinders their implementation on resource-constrained edge devices. Spiking Neural Networks (SNNs) offer an energy-efficient alternative due to their event-driven nature; however, their integration with continuous Self-Supervised Learning (SSL) representations is fundamentally challenged by distribution mismatch, where high-dynamic-range embeddings degrade the information coding capacity of threshold-based neurons. To resolve this, we propose Prompt-Tuned Spiking Neural Networks (PTS-SNN), a parameter-efficient neuromorphic adaptation framework that aligns frozen SSL backbones with spiking dynamics. Specifically, we introduce a Temporal Shift Spiking Encoder to capture local temporal dependencies via parameter-free channel shifts, establishing a stable feature basis. To bridge the domain gap, we devise a Context-Aware Membrane Potential Calibration strategy. This mechanism leverages a Spiking Sparse Linear Attention module to aggregate global semantic context into learnable soft prompts, which dynamically regulate the bias voltages of Parametric Leaky Integrate-and-Fire (PLIF) neurons. This regulation effectively centers the heterogeneous input distribution within the responsive firing range, mitigating functional silence or saturation. Extensive experiments on five multilingual datasets (e.g., IEMOCAP, CASIA, EMODB) demonstrate that PTS-SNN achieves 73.34\% accuracy on IEMOCAP, comparable to competitive Artificial Neural Networks (ANNs), while requiring only 1.19M trainable parameters and 0.35 mJ inference energy per sample.

cs.AI↗
arXiv2026

OFDM Enabled Over-the-Air Computation Systems with Two-Dimensional Fluid Antennas

Fluid antenna system (FAS) is able to exploit spatial degrees of freedom (DoFs) in wireless channels. In this letter, to exploit spatial DoFs in frequency-selective environments, we investigate an orthogonal frequency division multiplexing enabled over-the-air computation system, where the access point is equipped with a two-dimensional FAS to enhance performance. We solve the computation mean square error (MSE) minimization problem by transforming the original problem into transmit precoders optimization problem and antenna positions optimization along with receive combiners optimization problem. The latter is solved via a majorization-minimization approach combined with sequential optimization. Numerical results confirm that the proposed scheme achieves MSE reduction over the scheme with fixed position antennas.

cs.IT↗
arXiv2026

Steering LLMs via Scalable Interactive Oversight

As Large Language Models increasingly automate complex, long-horizon tasks such as \emph{vibe coding}, a supervision gap has emerged. While models excel at execution, users often struggle to guide them effectively due to insufficient domain expertise, the difficulty of articulating precise intent, and the inability to reliably validate complex outputs. It presents a critical challenge in scalable oversight: enabling humans to responsibly steer AI systems on tasks that surpass their own ability to specify or verify. To tackle this, we propose Scalable Interactive Oversight, a framework that decomposes complex intent into a recursive tree of manageable decisions to amplify human supervision. Rather than relying on open-ended prompting, our system elicits low-burden feedback at each node and recursively aggregates these signals into precise global guidance. Validated in web development task, our framework enables non-experts to produce expert-level Product Requirement Documents, achieving a 54\% improvement in alignment. Crucially, we demonstrate that this framework can be optimized via Reinforcement Learning using only online user feedback, offering a practical pathway for maintaining human control as AI scales.

cs.AI↗
arXiv2026

Cognitive algorithms and systems of episodic memory, semantic memory and their learnings

Declarative memory, the memory that can be "declared" in words or languages, is made up of two dissociated parts: episodic memory and semantic memory. This dissociation has its neuroanatomical basis episodic memory is mostly associated with the hippocampus and semantic memory with the neocortex. The two memories, on the other hand, are closely related. Lesions in the hippocampus often result in various impairments of explicit memory, e.g., anterograde, retrograde and developmental amnesias, and semantic learning deficit. These impairments provide opportunities for us to understand how the two memories may be acquired, stored and organized. This chapter reviews several cognitive systems that are centered to mimic explicit memory, and other systems that are neuroanatomically based and are implemented to simulate those memory impairments mentioned above. This review includes: the structures of the computational systems, their learning rules, and their simulations of memory acquisition and impairments.

q-bio.NC↗
arXiv2026

Swordsman: Entropy-Driven Adaptive Block Partition for Efficient Diffusion Language Models

Block-wise decoding effectively improves the inference speed and quality in diffusion language models (DLMs) by combining inter-block sequential denoising and intra-block parallel unmasking. However, existing block-wise decoding methods typically partition blocks in a rigid and fixed manner, which inevitably fragments complete semantic or syntactic constituents, leading to suboptimal performance. Inspired by the entropy reduction hypothesis (ERH), we recognize that constituent boundaries offer greater opportunities for uncertainty reduction, which motivates us to employ entropy analysis for identifying constituent boundaries. Therefore, we propose Swordsman, an entropy-driven adaptive block-wise decoding framework for DLMs. Swordsman adaptively partitions blocks by identifying entropy shifts between adjacent tokens to better align with semantic or syntactic constituent boundaries. In addition, Swordsman dynamically adjusts unmasking thresholds conditioned on the real-time unmasking status within a block, further improving both efficiency and stability. As a training-free framework, supported by KV Cache, Swordsman demonstrates state-of-the-art performance across extensive evaluations.

cs.CL↗
arXiv2026

Outcome Accuracy is Not Enough: Aligning the Reasoning Process of Reward Models

Generative Reward Models (GenRMs) and LLM-as-a-Judge exhibit deceptive alignment by producing correct judgments for incorrect reasons, as they are trained and evaluated to prioritize Outcome Accuracy, which undermines their ability to generalize during RLHF. We introduce Rationale Consistency, a fine-grained metric that quantifies the alignment between the model's reasoning process and human judgment. Our evaluation of frontier models reveals that rationale consistency effectively discriminates among state-of-the-art models and detects deceptive alignment, while outcome accuracy falls short in both respects. To mitigate this gap, we introduce a hybrid signal that combines rationale consistency with outcome accuracy for GenRM training. Our training method achieves state-of-the-art performance on RM-Bench (87.1%) and JudgeBench (82%), surpassing outcome-only baselines by an average of 5%. Using RM during RLHF, our method effectively improves performance as demonstrated on Arena Hard v2, notably yielding a 7% improvement in creative writing tasks. Further analysis confirms that our method escapes the deceptive alignment trap, effectively reversing the decline in rationale consistency observed in outcome-only training.

cs.CL↗
arXiv2026

A computational account of dreaming: learning and memory consolidation

A number of studies have concluded that dreaming is mostly caused by randomly arriving internal signals because "dream contents are random impulses", and argued that dream sleep is unlikely to play an important part in our intellectual capacity. On the contrary, numerous functional studies have revealed that dream sleep does play an important role in our learning and other intellectual functions. Specifically, recent studies have suggested the importance of dream sleep in memory consolidation, following the findings of neural replaying of recent waking patterns in the hippocampus. The randomness has been the hurdle that divides dream theories into either functional or functionless. This study presents a cognitive and computational model of dream process. This model is simulated to perform the functions of learning and memory consolidation, which are two most popular dream functions that have been proposed. The simulations demonstrate that random signals may result in learning and memory consolidation. Thus, dreaming is proposed as a continuation of brain's waking activities that processes signals activated spontaneously and randomly from the hippocampus. The characteristics of the model are discussed and found in agreement with many characteristics concluded from various empirical studies.

q-bio.NC↗
arXiv2026

CL-bench: A Benchmark for Context Learning

Current language models (LMs) excel at reasoning over prompts using pre-trained knowledge. However, real-world tasks are far more complex and context-dependent: models must learn from task-specific context and leverage new knowledge beyond what is learned during pre-training to reason and resolve tasks. We term this capability context learning, a crucial ability that humans naturally possess but has been largely overlooked. To this end, we introduce CL-bench, a real-world benchmark consisting of 500 complex contexts, 1,899 tasks, and 31,607 verification rubrics, all crafted by experienced domain experts. Each task is designed such that the new content required to resolve it is contained within the corresponding context. Resolving tasks in CL-bench requires models to learn from the context, ranging from new domain-specific knowledge, rule systems, and complex procedures to laws derived from empirical data, all of which are absent from pre-training. This goes far beyond long-context tasks that primarily test retrieval or reading comprehension, and in-context learning tasks, where models learn simple task patterns via instructions and demonstrations. Our evaluations of ten frontier LMs find that models solve only 17.2% of tasks on average. Even the best-performing model, GPT-5.1, solves only 23.7%, revealing that LMs have yet to achieve effective context learning, which poses a critical bottleneck for tackling real-world, complex context-dependent tasks. CL-bench represents a step towards building LMs with this fundamental capability, making them more intelligent and advancing their deployment in real-world scenarios.

cs.CL↗
arXiv2026

AdNanny: One Reasoning LLM for All Offline Ads Recommendation Tasks

Large Language Models (LLMs) have shown strong capabilities in Natural Language Understanding and Generation, but deploying them directly in online advertising systems is often impractical due to strict millisecond-level latency constraints. This has motivated the use of LLMs offline to improve retrieval, ranking, and recommendation models. Existing solutions typically fine-tune separate LLMs for individual tasks such as query-ad relevance labeling, keyword-based query generation, and user profiling. This results in redundant models, high maintenance cost, and limited performance gains despite substantial overlap in domain knowledge and reasoning patterns. We introduce AdNanny, a unified reasoning-centric LLM that serves as a shared backbone for offline advertising tasks. AdNanny is obtained by fine-tuning a public 671B-parameter DeepSeek-R1 checkpoint using a scalable training system that supports hybrid dense-MoE parallelism. We construct reasoning-augmented corpora that pair structured supervision with step-by-step natural language explanations. A multi-task supervised fine-tuning stage with adaptive reweighting enables AdNanny to handle diverse labeling and generation tasks in a consistent reasoning format. This is followed by reinforcement learning using downstream advertising metrics to align model behavior with online retrieval and ranking objectives. AdNanny is deployed in production within Bing Ads, where it significantly reduces manual labeling effort and improves accuracy across multiple offline tasks. By consolidating many task-specific models into a single reasoning-centric foundation model, AdNanny provides a scalable and cost-effective solution for large-scale advertising systems.

cs.SE↗
arXiv2026

Towards Fair Large Language Model-based Recommender Systems without Costly Retraining

Large Language Models (LLMs) have revolutionized Recommender Systems (RS) through advanced generative user modeling. However, LLM-based RS (LLM-RS) often inadvertently perpetuates bias present in the training data, leading to severe fairness issues. Addressing these fairness problems in LLM-RS faces two significant challenges. 1) Existing debiasing methods, designed for specific bias types, lack the generality to handle diverse or emerging biases in real-world applications. 2) Debiasing methods relying on retraining are computationally infeasible given the massive parameter scale of LLMs. To overcome these challenges, we propose FUDLR (Fast Unified Debiasing for LLM-RS). The core idea is to reformulate the debiasing problem as an efficient machine unlearning task with two stages. First, FUDLR identifies bias-inducing samples to unlearn through a novel bias-agnostic mask, optimized to balance fairness improvement with accuracy preservation. Its bias-agnostic design allows adaptability to various or co-existing biases simply by incorporating different fairness metrics. Second, FUDLR performs efficient debiasing by estimating and removing the influence of identified samples on model parameters. Extensive experiments demonstrate that FUDLR effectively and efficiently improves fairness while preserving recommendation accuracy, offering a practical path toward socially responsible LLM-RS. The code and data are available at https://github.com/JinLi-i/FUDLR.

cs.IR↗
arXiv2026

FaithSCAN: Model-Driven Single-Pass Hallucination Detection for Faithful Visual Question Answering

Faithfulness hallucinations in VQA occur when vision-language models produce fluent yet visually ungrounded answers, severely undermining their reliability in safety-critical applications. Existing detection methods mainly fall into two categories: external verification approaches relying on auxiliary models or knowledge bases, and uncertainty-driven approaches using repeated sampling or uncertainty estimates. The former suffer from high computational overhead and are limited by external resource quality, while the latter capture only limited facets of model uncertainty and fail to sufficiently explore the rich internal signals associated with the diverse failure modes. Both paradigms thus have inherent limitations in efficiency, robustness, and detection performance. To address these challenges, we propose FaithSCAN: a lightweight network that detects hallucinations by exploiting rich internal signals of VLMs, including token-level decoding uncertainty, intermediate visual representations, and cross-modal alignment features. These signals are fused via branch-wise evidence encoding and uncertainty-aware attention. We also extend the LLM-as-a-Judge paradigm to VQA hallucination and propose a low-cost strategy to automatically generate model-dependent supervision signals, enabling supervised training without costly human labels while maintaining high detection accuracy. Experiments on multiple VQA benchmarks show that FaithSCAN significantly outperforms existing methods in both effectiveness and efficiency. In-depth analysis shows hallucinations arise from systematic internal state variations in visual perception, cross-modal reasoning, and language decoding. Different internal signals provide complementary diagnostic cues, and hallucination patterns vary across VLM architectures, offering new insights into the underlying causes of multimodal hallucinations.

cs.CV↗
arXiv2026

ChartE$^{3}$: A Comprehensive Benchmark for End-to-End Chart Editing

Charts are a fundamental visualization format for structured data analysis. Enabling end-to-end chart editing according to user intent is of great practical value, yet remains challenging due to the need for both fine-grained control and global structural consistency. Most existing approaches adopt pipeline-based designs, where natural language or code serves as an intermediate representation, limiting their ability to faithfully execute complex edits. We introduce ChartE$^{3}$, an End-to-End Chart Editing benchmark that directly evaluates models without relying on intermediate natural language programs or code-level supervision. ChartE$^{3}$ focuses on two complementary editing dimensions: local editing, which involves fine-grained appearance changes such as font or color adjustments, and global editing, which requires holistic, data-centric transformations including data filtering and trend line addition. ChartE$^{3}$ contains over 1,200 high-quality samples constructed via a well-designed data pipeline with human curation. Each sample is provided as a triplet of a chart image, its underlying code, and a multimodal editing instruction, enabling evaluation from both objective and subjective perspectives. Extensive benchmarking of state-of-the-art multimodal large language models reveals substantial performance gaps, particularly on global editing tasks, highlighting critical limitations in current end-to-end chart editing capabilities.

cs.CV↗
arXiv2026

Hint: hierarchical inter-frame correlation for one-shot point cloud sequence compression

Deep learning has demonstrated strong capability in compressing point clouds. Within this area, entropy modeling for lossless compression is widely investigated. However, most methods rely solely on parent/sibling contexts and level-wise autoregression, which suffers from decoding latency on the order of 10^1-10^2 seconds. We propose HINT, a method that integrates temporal and spatial correlation for sequential point cloud compression. Specifically, it first uses a two-stage temporal feature extraction: (i) a parent-level existence map and (ii) a child-level neighborhood lookup in the previous frame. These cues are fused with the spatial features via element-wise addition and encoded with a group-wise strategy. Experimental results show that HINT achieves encoding and decoding time at 105 ms and 140 ms, respectively, equivalent to 49.6x and 21.6x acceleration in comparison with G-PCC, while achieving up to 43.6% bitrate reduction and consistently outperforming the spatial-only baseline (RENO).

eess.IV↗
arXiv2026

Pretrain Value, Not Reward: Decoupled Value Policy Optimization

In this paper, we explore how directly pretraining a value model simplifies and stabilizes reinforcement learning from human feedback (RLHF). In reinforcement learning, value estimation is the key to policy optimization, distinct from reward supervision. The value function predicts the \emph{return-to-go} of a partial answer, that is, how promising the partial answer is if it were continued to completion. In RLHF, however, the standard pipeline first pretrains a reward model and then learns a value function online, even though no new reward signals are available once preference data is collected. This makes critic learning redundant, as the process of training a reward model and then deriving a value model is informationally equivalent to directly pretraining a value model. Importantly, this requires no additional supervision, and our value model is trained on exactly the same data used for reward modeling. Building on this insight, we introduce \emph{Decoupled Value Policy Optimization} (DVPO), a framework that pretrains a \emph{Global Value Model} (GVM) offline and freezes it as a universal critic for policy learning. The GVM provides stable, fine-grained credit assignment without critic drift or trajectory sampling. Experiments across MT-Bench, Alpaca-Eval, and Arena-Hard demonstrate that DVPO matches or surpasses state-of-the-art RLHF methods. These results highlight RLHF can be reframed as policy-only optimization guided by a single pretrained value model.

cs.LG↗
arXiv2026

Shapley Features for Robust Signal Prediction in Tactile Internet

The Tactile Internet (TI) requires ultra-low latency and reliable haptic signal transmission, yet packet loss and delay remain unresolved challenges. We present a novel prediction framework that integrates Gaussian Processes (GP) with a ResNet-based Neural Network, where GP acts as an oracle to recover signals lost or heavily delayed. To further optimize performance, we introduce Shapley Feature Values (SFV), a principled feature selection mechanism that isolates the most informative inputs for prediction. This GP+SFV framework achieves 95.72% accuracy, surpassing the state-of-the-art LeFo method by 11.1%, while simultaneously relaxing TI's rigid delay constraints. Beyond accuracy, SFV operates as a modular accelerator: when paired with LeFo, it reduces inference time by 27%, and when paired with GP, by 72%. These results establish GP+SFV as both a high-accuracy and high-efficiency solution, paving the way for practical and reliable haptic communications in TI systems.

eess.SP↗
arXiv2026

Discrete Mode Decomposition Meets Shapley Value: Robust Signal Prediction in Tactile Internet

Tactile Internet (TI) requires ultra-low latency and high reliability to ensure stability and transparency in touch-enabled teleoperation. However, variable delays and packet loss present significant challenges to maintaining immersive haptic communication. To address this, we propose a predictive framework that integrates Discrete Mode Decomposition (DMD) with Shapley Mode Value (SMV) for accurate and timely haptic signal prediction. DMD decomposes haptic signals into interpretable intrinsic modes, while SMV evaluates each mode's contribution to prediction accuracy, which is well-aligned with the goal-oriented semantic communication. Integrating SMV with DMD further accelerates inference, enabling efficient communication and smooth teleoperation even under adverse network conditions. Extensive experiments show that DMD+SMV, combined with a Transformer architecture, outperforms baseline methods significantly. It achieves 98.9% accuracy for 1-sample prediction and 92.5% for 100-sample prediction, as well as extremely low inference latency: 0.056 ms and 2 ms, respectively. These results demonstrate that the proposed framework has strong potential to ease the stringent latency and reliability requirements of TI without compromising performance, highlighting its feasibility for real-world deployment in TI systems.

eess.SP↗
arXiv2026

GTR-Mamba: Geometry-to-Tangent Routing Mamba for Hyperbolic POI Recommendation

Next Point-of-Interest (POI) recommendation is a critical task in modern Location-Based Social Networks (LBSNs), aiming to model the complex decision-making process of human mobility to provide personalized recommendations for a user's next check-in location. Existing hyperbolic POI recommendation models, predominantly based on rotations and graph representations, have been extensively investigated. Although hyperbolic geometry has proven superior in representing hierarchical data with low distortion, current hyperbolic sequence models typically rely on performing recurrence via expensive Möbius operations directly on the manifold. This incurs prohibitive computational costs and numerical instability, rendering them ill-suited for trajectory modeling. To resolve this conflict between geometric representational power and sequential efficiency, we propose GTR-Mamba, a novel framework featuring Geometry-to-Tangent Routing. GTR-Mamba strategically routes complex state transitions to the computationally efficient Euclidean tangent space. Crucially, instead of a static approximation, we introduce a Parallel Transport (PT) mechanism that dynamically aligns tangent spaces along the trajectory. This ensures geometric consistency across recursive updates, effectively bridging the gap between the curved manifold and linear tangent operations. This process is orchestrated by an exogenous spatio-temporal channel, which explicitly modulates the SSM discretization parameters. Extensive experiments on three real-world datasets demonstrate that GTR-Mamba consistently outperforms state-of-the-art baselines in next POI recommendation.

cs.AI↗