arXiv subjects

Shida Wang

Publications and source records attributed to Shida Wang.

At least 19 recordsLinked to original sources

arXiv2026

VISTA: Verifier-Informed Student-to-Teacher Adaptation for On-Policy Self-Distillation

On-policy self-distillation (OPSD) improves reasoning by training a problem-only student on its own rollouts using dense token-level supervision from a privileged teacher that also sees a reference solution. However, standard OPSD treats the teacher distribution as a fixed target along the student's rollout and updates only the student %, although -- even though privileged conditioning does not guarantee that the teacher always provides the most appropriate target for problem-only reasoning. This one-way supervision can therefore misdirect the student when the teacher distribution is misaligned with valid student reasoning. We therefore introduce Verifier-Informed Student-to-Teacher Adaptation (VISTA), which preserves the standard OPSD student update while using outcome-verified rollouts to adapt the teacher toward the student distribution. Within each verified rollout, VISTA further restricts this adaptation to the top-$k$ positions with the largest teacher--student KL divergence. Notably, VISTA reuses the rollout and loss function from standard OPSD, introducing no additional sampling or separate reward objective. Across AIME24, AIME25, and HMMT25 with Qwen3 models at 1.7B, 4B, and 8B, VISTA achieves the highest Avg@12 at every scale, improving over OPSD by $0.6$, $0.7$, and $2.1$ points, respectively. These results demonstrate the value of student supervision from outcome-verified rollouts and highlight student-to-teacher adaptation as a promising direction for OPSD.

cs.LG↗

arXiv2026

LOCUS: Local Visual Cue Search for Enhancing Fine-Grained Perception in Multimodal Large Language Models

Multimodal Large Language Models (MLLMs) remain unreliable on fine-grained visual perception, even when high-resolution inputs preserve the necessary local details. We identify this limitation as visual context rot: decisive evidence may exist in the full image, yet fail to be reliably selected and used amid redundant visual context. We propose LOCUS (LOcal visual CUe Search), a training framework that teaches MLLMs to internalize local evidence search through a verifiable proxy task. During training, LOCUS provides a local crop as a visual cue and optimizes the model to recover its spatial support in the full image using an IoU-based reward. The visual cue is used only during training, leaving the standard image-question inference interface unchanged. Experiments across fine-grained perception, hallucination, general understanding, and reasoning benchmarks show that LOCUS improves localization-sensitive visual understanding while preserving broad capabilities. Attention analyses further indicate stronger focus on task-relevant evidence regions, suggesting that training-time visual cue search provides an effective route to internalized fine-grained evidence selection.

cs.CV↗

arXiv2026

Dynamic Token Compression for Efficient Video Understanding through Reinforcement Learning

Multimodal Large Language Models have demonstrated remarkable capabilities in video understanding, yet face prohibitive computational costs and performance degradation from ''context rot'' due to massive visual token redundancy. Existing compression strategies typically rely on heuristics or fixed transformations that are often decoupled from the downstream task objectives, limiting their adaptability and effectiveness. To address this, we propose SCORE (Surprise-augmented token COmpression via REinforcement learning), a unified framework that learns an adaptive token compression policy. SCORE introduces a lightweight policy network conditioned on a surprise-augmented state representation that incorporates inter-frame residuals to explicitly capture temporal dynamics and motion saliency. We optimize this policy using a group-wise reinforcement learning scheme with a split-advantage estimator, stabilized by a two-stage curriculum transferring from static pseudo-videos to real dynamic videos. Extensive experiments on diverse video understanding benchmarks demonstrate that SCORE significantly outperforms state-of-the-art baselines. Notably, SCORE achieves a 16x prefill speedup while preserving 99.5% of original performance at a 10% retention ratio, offering a scalable solution for efficient long-form video understanding.

cs.CV↗

arXiv2025

DiG: Differential Grounding for Enhancing Fine-Grained Perception in Multimodal Large Language Model

Multimodal Large Language Models have achieved impressive performance on a variety of vision-language tasks, yet their fine-grained visual perception and precise spatial reasoning remain limited. In this work, we introduce DiG (Differential Grounding), a novel proxy task framework where MLLMs learn fine-grained perception by identifying and localizing all differences between similar image pairs without prior knowledge of their number. To support scalable training, we develop an automated 3D rendering-based data generation pipeline that produces high-quality paired images with fully controllable discrepancies. To address the sparsity of difference signals, we further employ curriculum learning that progressively increases complexity from single to multiple differences, enabling stable optimization. Extensive experiments demonstrate that DiG significantly improves model performance across a variety of visual perception benchmarks and that the learned fine-grained perception skills transfer effectively to standard downstream tasks, including RefCOCO, RefCOCO+, RefCOCOg, and general multimodal perception benchmarks. Our results highlight differential grounding as a scalable and robust approach for advancing fine-grained visual reasoning in MLLMs.

cs.CV↗

arXiv2025

FPEdit: Robust LLM Fingerprinting through Localized Parameter Editing

Large language models represent significant investments in computation, data, and engineering expertise, making them extraordinarily valuable intellectual assets. Nevertheless, these AI assets remain vulnerable to unauthorized redistribution and commercial exploitation through fine-tuning or black-box deployment. Current fingerprinting approaches face a fundamental trade-off: intrinsic methods require full parameter access, while backdoor-based techniques employ statistically anomalous triggers easily detected and filtered by adversaries. To address these limitations, we introduce FPEdit, a novel framework that leverages knowledge editing to inject semantically coherent natural language fingerprints through sparse, targeted modifications to model weights. Our approach introduces Promote-Suppress Value Vector Optimization, which simultaneously enhances target token likelihood while suppressing competing tokens, ensuring robust fingerprint integration without degrading core model functionality. Extensive experiments show that FPEdit achieves 94-100% fingerprint retention under both full-parameter fine-tuning and parameter-efficient adaptation, while preserving performance on downstream benchmarks. Moreover, FPEdit remains robust under quantization, pruning, and stochastic decoding, and can embed 10 fingerprint pairs into LLaMA2-7B in under 2 minutes using less than 30 GB of GPU memory, which represents a substantial reduction in resource requirements. These advances establish FPEdit as the first fingerprinting approach to simultaneously achieve robustness against adaptation, resistance to detection, and preservation of model utility, thereby providing a minimally invasive solution for reliable provenance verification of large language models in adversarial deployment scenarios.

cs.CR↗

arXiv2025

Numerical Investigation of Sequence Modeling Theory using Controllable Memory Functions

The evolution of sequence modeling architectures, from recurrent neural networks and convolutional models to Transformers and structured state-space models, reflects ongoing efforts to address the diverse temporal dependencies inherent in sequential data. Despite this progress, systematically characterizing the strengths and limitations of these architectures remains a fundamental challenge. In this work, we propose a synthetic benchmarking framework to evaluate how effectively different sequence models capture distinct temporal structures. The core of this approach is to generate synthetic targets, each characterized by a parametric memory function $\rho(s, \alpha)$ and a controllable parameter $\alpha$ that determines the temporal strength. This setup allows us to produce a continuum of tasks that vary in temporal complexity, enabling fine-grained analysis of model behavior with respect to specific memory properties. We focus on four representative memory functions, each corresponding to a distinct class of temporal structures: exponential and polynomial functions for decay dynamics, impulse functions for long-range dependencies, and Airy functions for sparsity patterns. Experiments on several sequence modeling architectures confirm existing theoretical insights and reveal new findings regarding approximation capabilities, optimization dynamics, and architectural trade-offs. These results demonstrate the effectiveness of the proposed method in advancing theoretical understanding and highlight the importance of using controllable targets with clearly defined structures for evaluating sequence modeling architectures.

cs.LG↗

arXiv2025

Convergence rates of regularized quasi-Newton methods without strong convexity

In this paper, we study convergence rates of the cubic regularized proximal quasi-Newton method (\csr) for solving non-smooth additive composite problems that satisfy the so-called Kurdyka-\L ojasiewicz (K\L ) property with respect to some desingularization function $\phi$ rather than strong convexity. After a number of iterations $k_0$, Cubic SR1 PQN exhibits non-asymptotic explicit super-linear convergence rates for any $k\geq k_0$. In particular, when $\phi(t)=ct^{1/2}$, Cubic SR1 PQN has a convergence rate of order $\left(\frac{C}{(k-k_0)^{1/2}}\right)^{(k-k_0)/2}$, where $k$ is the number of iterations and $C>0$ is a constant. For the special case, i.e. functions which satisfy \L ojasiewicz inequality, the rate becomes global and non-asymptotic. This work presents, for the first time, non-asymptotic explicit convergence rates of regularized (proximal) SR1 quasi-Newton methods applied to non-convex non-smooth problems with K\L\ property. Actually, the rates are novel even in the smooth non-convex case. Notably, we achieve this without employing line search or trust region strategies, without assuming the Dennis-Mor\'e condition, without any assumptions on quasi-Newton metrics and without assuming strong convexity. Furthermore, for convex problems, we focus on a more tractable gradient regularized quasi-Newton method (Grad SR1 PQN) which can achieve results similar to those obtained with cubic regularization. We also demonstrate, for the first time, the non-asymptotic super-linear convergence rate of Grad SR1 PQN for solving convex problems with the help of the \L ojasiewicz inequality instead of strong convexity.

math.OC↗

arXiv2025

Learning Relational Tabular Data without Shared Features

Learning relational tabular data has gained significant attention recently, but most studies focus on single tables, overlooking the potential of cross-table learning. Cross-table learning, especially in scenarios where tables lack shared features and pre-aligned data, offers vast opportunities but also introduces substantial challenges. The alignment space is immense, and determining accurate alignments between tables is highly complex. We propose Latent Entity Alignment Learning (Leal), a novel framework enabling effective cross-table training without requiring shared features or pre-aligned data. Leal operates on the principle that properly aligned data yield lower loss than misaligned data, a concept embodied in its soft alignment mechanism. This mechanism is coupled with a differentiable cluster sampler module, ensuring efficient scaling to large relational tables. Furthermore, we provide a theoretical proof of the cluster sampler's approximation capacity. Extensive experiments on five real-world and five synthetic datasets show that Leal achieves up to a 26.8% improvement in predictive performance compared to state-of-the-art methods, demonstrating its effectiveness and scalability.

cs.LG↗

arXiv2024

Global non-asymptotic super-linear convergence rates of regularized proximal quasi-Newton methods on non-smooth composite problems

In this paper, we propose two regularized proximal quasi-Newton methods with symmetric rank-1 update of the metric (SR1 quasi-Newton) to solve non-smooth convex additive composite problems. Both algorithms avoid using line search or other trust region strategies. For each of them, we prove a super-linear convergence rate that is independent of the initialization of the algorithm. The cubic regularized method achieves a rate of order $\left(\frac{C}{N^{1/2}}\right)^{N/2}$, where $N$ is the number of iterations and $C$ is some constant, and the other gradient regularized method shows a rate of the order $\left(\frac{C}{N^{1/4}}\right)^{N/2}$. To the best of our knowledge, these are the first global non-asymptotic super-linear convergence rates for regularized quasi-Newton methods and regularized proximal quasi-Newton methods. The theoretical properties are also demonstrated in two applications from machine learning.

math.OC↗

arXiv2024

LongSSM: On the Length Extension of State-space Models in Language Modelling

In this paper, we investigate the length-extension of state-space models (SSMs) in language modeling. Length extension involves training models on short sequences and testing them on longer ones. We show that state-space models trained with zero hidden states initialization have difficulty doing length extension. We explain this difficulty by pointing out the length extension is equivalent to polynomial extrapolation. Based on the theory, we propose a simple yet effective method - changing the hidden states initialization scheme - to improve the length extension. Moreover, our method shows that using long training sequence length is beneficial but not necessary to length extension. Changing the hidden state initialization enables the efficient training of long-memory model with a smaller training context length.

cs.CL↗

arXiv2024

A Quasi-Newton Primal-Dual Algorithm with Line Search

Quasi-Newton methods refer to a class of algorithms at the interface between first and second order methods. They aim to progress as substantially as second order methods per iteration, while maintaining the computational complexity of first order methods. The approximation of second order information by first order derivatives can be expressed as adopting a variable metric, which for (limited memory) quasi-Newton methods is of type ``identity $\pm$ low rank''. This paper continues the effort to make these powerful methods available for non-smooth systems occurring, for example, in large scale Machine Learning applications by exploiting this special structure. We develop a line search variant of a recently introduced quasi-Newton primal-dual algorithm, which adds significant flexibility, admits larger steps per iteration, and circumvents the complicated precalculation of a certain operator norm. We prove convergence, including convergence rates, for our proposed method and outperform related algorithms in a large scale image deblurring application.

math.OC↗

arXiv2024

Integrating Deep Learning and Synthetic Biology: A Co-Design Approach for Enhancing Gene Expression via N-terminal Coding Sequences

N-terminal coding sequence (NCS) influences gene expression by impacting the translation initiation rate. The NCS optimization problem is to find an NCS that maximizes gene expression. The problem is important in genetic engineering. However, current methods for NCS optimization such as rational design and statistics-guided approaches are labor-intensive yield only relatively small improvements. This paper introduces a deep learning/synthetic biology co-designed few-shot training workflow for NCS optimization. Our method utilizes k-nearest encoding followed by word2vec to encode the NCS, then performs feature extraction using attention mechanisms, before constructing a time-series network for predicting gene expression intensity, and finally a direct search algorithm identifies the optimal NCS with limited training data. We took green fluorescent protein (GFP) expressed by Bacillus subtilis as a reporting protein of NCSs, and employed the fluorescence enhancement factor as the metric of NCS optimization. Within just six iterative experiments, our model generated an NCS (MLD62) that increased average GFP expression by 5.41-fold, outperforming the state-of-the-art NCS designs. Extending our findings beyond GFP, we showed that our engineered NCS (MLD62) can effectively boost the production of N-acetylneuraminic acid by enhancing the expression of the crucial rate-limiting GNA1 gene, demonstrating its practical utility. We have open-sourced our NCS expression database and experimental procedures for public use.

q-bio.QM↗

arXiv2023

StableSSM: Alleviating the Curse of Memory in State-space Models through Stable Reparameterization

In this paper, we investigate the long-term memory learning capabilities of state-space models (SSMs) from the perspective of parameterization. We prove that state-space models without any reparameterization exhibit a memory limitation similar to that of traditional RNNs: the target relationships that can be stably approximated by state-space models must have an exponential decaying memory. Our analysis identifies this "curse of memory" as a result of the recurrent weights converging to a stability boundary, suggesting that a reparameterization technique can be effective. To this end, we introduce a class of reparameterization techniques for SSMs that effectively lift its memory limitations. Besides improving approximation capabilities, we further illustrate that a principled choice of reparameterization scheme can also enhance optimization stability. We validate our findings using synthetic datasets, language models and image classifications.

cs.LG↗

arXiv2023

State-space Models with Layer-wise Nonlinearity are Universal Approximators with Exponential Decaying Memory

State-space models have gained popularity in sequence modelling due to their simple and efficient network structures. However, the absence of nonlinear activation along the temporal direction limits the model's capacity. In this paper, we prove that stacking state-space models with layer-wise nonlinear activation is sufficient to approximate any continuous sequence-to-sequence relationship. Our findings demonstrate that the addition of layer-wise nonlinear activation enhances the model's capacity to learn complex sequence patterns. Meanwhile, it can be seen both theoretically and empirically that the state-space models do not fundamentally resolve the issue of exponential decaying memory. Theoretical results are justified by numerical verifications.

cs.LG↗

arXiv2023

Efficient Hyperdimensional Computing

Hyperdimensional computing (HDC) is a method to perform classification that uses binary vectors with high dimensions and the majority rule. This approach has the potential to be energy-efficient and hence deemed suitable for resource-limited platforms due to its simplicity and massive parallelism. However, in order to achieve high accuracy, HDC sometimes uses hypervectors with tens of thousands of dimensions. This potentially negates its efficiency advantage. In this paper, we examine the necessity of such high dimensions and conduct a detailed theoretical analysis of the relationship between hypervector dimensions and accuracy. Our results demonstrate that as the dimension of the hypervectors increases, the worst-case/average-case HDC prediction accuracy with the majority rule decreases. Building on this insight, we develop HDC models that use binary hypervectors with dimensions orders of magnitude lower than those of state-of-the-art HDC models while maintaining equivalent or even improved accuracy and efficiency. For instance, on the MNIST dataset, we achieve 91.12% HDC accuracy in image classification with a dimension of only 64. Our methods perform operations that are only 0.35% of other HDC models with dimensions of 10,000. Furthermore, we evaluate our methods on ISOLET, UCI-HAR, and Fashion-MNIST datasets and investigate the limits of HDC computing.

cs.LG↗

arXiv2023

HyperSNN: A new efficient and robust deep learning model for resource constrained control applications

In light of the increasing adoption of edge computing in areas such as intelligent furniture, robotics, and smart homes, this paper introduces HyperSNN, an innovative method for control tasks that uses spiking neural networks (SNNs) in combination with hyperdimensional computing. HyperSNN substitutes expensive 32-bit floating point multiplications with 8-bit integer additions, resulting in reduced energy consumption while enhancing robustness and potentially improving accuracy. Our model was tested on AI Gym benchmarks, including Cartpole, Acrobot, MountainCar, and Lunar Lander. HyperSNN achieves control accuracies that are on par with conventional machine learning methods but with only 1.36% to 9.96% of the energy expenditure. Furthermore, our experiments showed increased robustness when using HyperSNN. We believe that HyperSNN is especially suitable for interactive, mobile, and wearable devices, promoting energy-efficient and robust system design. Furthermore, it paves the way for the practical implementation of complex algorithms like model predictive control (MPC) in real-world industrial scenarios.

cs.RO↗

arXiv2023

Improve Long-term Memory Learning Through Rescaling the Error Temporally

This paper studies the error metric selection for long-term memory learning in sequence modelling. We examine the bias towards short-term memory in commonly used errors, including mean absolute/squared error. Our findings show that all temporally positive-weighted errors are biased towards short-term memory in learning linear functionals. To reduce this bias and improve long-term memory learning, we propose the use of a temporally rescaled error. In addition to reducing the bias towards short-term memory, this approach can also alleviate the vanishing gradient issue. We conduct numerical experiments on different long-memory tasks and sequence models to validate our claims. Numerical results confirm the importance of appropriate temporally rescaled error for effective long-term memory learning. To the best of our knowledge, this is the first work that quantitatively analyzes different errors' memory bias towards short-term memory in sequence modelling.

cs.LG↗

arXiv2023

Inverse Approximation Theory for Nonlinear Recurrent Neural Networks

We prove an inverse approximation theorem for the approximation of nonlinear sequence-to-sequence relationships using recurrent neural networks (RNNs). This is a so-called Bernstein-type result in approximation theory, which deduces properties of a target function under the assumption that it can be effectively approximated by a hypothesis space. In particular, we show that nonlinear sequence relationships that can be stably approximated by nonlinear RNNs must have an exponential decaying memory structure - a notion that can be made precise. This extends the previously identified curse of memory in linear RNNs into the general nonlinear setting, and quantifies the essential limitations of the RNN architecture for learning sequential relationships with long-term memory. Based on the analysis, we propose a principled reparameterization method to overcome the limitations. Our theoretical results are confirmed by numerical experiments. The code has been released in https://github.com/radarFudan/Curse-of-memory

cs.LG↗