Searcharxiv⌕ Search

arXiv subjects

Qitong Wang

Publications and source records attributed to Qitong Wang.

At least 19 recordsLinked to original sources
arXiv2026

Agentic Context Cracking: Token-Efficient Data Reasoning Agents via Adaptive Structuring of Unstructured Data

Valuable data remains embedded in unstructured sources: web pages, reports, contracts, filings, earnings calls, and PDFs. The big bet in enterprise AI is deploying LLM agents that reason over this data to answer complex questions for every knowledge worker. Agents can do this today, but at prohibitive cost. Each question repeatedly opens large documents to recover scattered evidence, consuming up to a million tokens. However, if the data were already structured, the same question would reduce to a cheap database lookup. For example, on FanOutQA benchmark, reasoning over an ideal pre-structured store is 28X cheaper, and the gap grows to orders of magnitude as questions fan out over more documents. Yet structuring everything in advance is not viable: documents hold vastly more possible structure than any workload will use, and the useful structure and documents are unknown until queries arrive. We propose agentic data cracking, a method that structures unstructured data adaptively and speculatively as a byproduct of reasoning itself. Structuring is adaptive because observed queries decide when it happens and what matters, and speculative because it goes beyond the current question. Whenever the agent opens a document to answer, a cracking sub-agent forks from the already-loaded context at marginal cost and extracts grounded structure likely to serve related future queries. Over time, an increasing share of queries is fully covered by structured data and answered without opening a document, keeping agentic accuracy at close to RAG cost. On FanOutQA, extended with merely one related question per test question, cracking cuts cost by 53% while preserving accuracy. Agentic data cracking is a first step toward next-generation data infrastructure for agentic reasoning over unstructured data: a shared substrate beneath the model where knowledge that reasoning already paid to uncover accumulates.

cs.AI↗
arXiv2026

Toward Low-Latency Vision-Language Models with Doubly-Correct Predictions in Egocentric Visual Understanding

The rapid rise of Vision-Language Models (VLMs) in egocentric visual understanding has made low-latency inference in human-robot collaborative (HRC) tasks increasingly critical. Weight pruning techniques developed for VLMs to shrink model size and computation can be readily applied to satisfy the efficiency demands of on-board processing and real-time interactive robotics. Moreover, safe human-robot interaction demands pruning strategies that preserve doubly-correct predictions; outputs must be both accurate and evidentially grounded to mitigate risks and ensure user trust. In this paper, we present a new study of VLM pruning through the lens of doubly-correct prediction. Our experiments surprisingly show that existing pruning methods often preserve the right evidence localization but undermine correct prediction. To address this, we propose a rationale-informed pruning strategy that better aligns evidence with decisions. Benchmark results on egocentric video datasets demonstrate that our method not only achieves the highest prediction accuracy but also outperforms existing approaches in attaining doubly-correct predictions. We aim to stimulate research on efficient and reliable VLMs, ensuring accuracy-driven advances align with the transparency, auditability, and safety required for responsible human-robot interaction and embodied intelligence.

cs.RO↗
arXiv2026

IceCache: Memory-efficient KV-cache Management for Long-Sequence LLMs

Key-Value (KV) cache plays a crucial role in accelerating inference in large language models (LLMs) by storing intermediate attention states and avoiding redundant computation during autoregressive generation. However, its memory footprint scales linearly with sequence length, often leading to severe memory bottlenecks on resource-constrained hardware. Prior work has explored offloading KV cache to the CPU while retaining only a subset on the GPU, but these approaches often rely on imprecise token selection and suffer performance degradation in long-generation tasks such as chain-of-thought reasoning. In this paper, we propose a novel KV cache management strategy, IceCache, which integrates semantic token clustering with PagedAttention. By organizing semantically related tokens into contiguous memory regions managed by a hierarchical, dynamically updatable data structure, our method enables more efficient token selection and better utilization of memory bandwidth during CPU-GPU transfers. Experimental results on LongBench show that, with a 256-token budget, IceCache maintains 99% of the original accuracy achieved by the full KV cache model. Moreover, compared to other offloading-based methods, IceCache attains competitive or even superior latency and accuracy while using only 25% of the KV cache token budget, demonstrating its effectiveness in long-sequence scenarios. The code is available on our project website at https://yuzhenmao.github.io/IceCache/.

cs.LG↗
arXiv2026

History-Conditioned Spatio-Temporal Visual Token Pruning for Efficient Vision-Language Navigation

Vision-Language Navigation (VLN) enables robots to follow natural-language instructions in visually grounded environments, serving as a key capability for embodied robotic systems. Recent Vision-Language-Action (VLA) models have demonstrated strong navigation performance, but their high computational cost introduces latency that limits real-time deployment. We propose a training-free spatio-temporal vision token pruning framework tailored to VLA-based VLN. We apply spatial token selection to the current view, alongside spatio-temporal compression for historical memories, enabling efficient long-horizon inference while reducing redundant computation. Leveraging attention-based token importance and query-guided spatio-temporal filtering, the proposed approach preserves navigation-relevant information without retraining or modifying pretrained models, allowing plug-and-play integration into existing VLA systems. Through experiments on standard VLN benchmarks, we confirm that our method significantly outperforms existing pruning strategies. It successfully preserves superior navigation accuracy under extreme pruning scenarios, all while maintaining the highly competitive inference efficiency. Real-world deployment on a Unitree Go2 quadruped robot further validates reliable and low-latency instruction-following navigation under practical robotic constraints. We hope this work helps bridge the gap between large-scale multimodal modeling and efficient, real-time embodied deployment in robotic navigation systems. Project Page: https://wqtwjt1996.github.io/publications/2026-vln.html

cs.RO↗
arXiv2026

When One Modality Rules Them All: Backdoor Modality Collapse in Multimodal Diffusion Models

While diffusion models have revolutionized visual content generation, their rapid adoption has underscored the critical need to investigate vulnerabilities, e.g., to backdoor attacks. In multimodal diffusion models, it is natural to expect that attacking multiple modalities simultaneously (e.g., text and image) would yield complementary effects and strengthen the overall backdoor. In this paper, we challenge this assumption by investigating the phenomenon of Backdoor Modality Collapse, a scenario where the backdoor mechanism degenerates to rely predominantly on a subset of modalities, rendering others redundant. To rigorously quantify this behavior, we introduce two novel metrics: Trigger Modality Attribution (TMA) and Cross-Trigger Interaction (CTI). Through extensive experiments across diverse training configurations in multimodal conditional diffusion, we consistently observe a ``winner-takes-all'' dynamic in backdoor behavior. Our results reveal that (1) attacks often collapse into subset-modality dominance, and (2) cross-modal interaction is negligible or even negative, contradicting the intuition of synergistic vulnerability. These findings highlight a critical blind spot in current assessments, suggesting that high attack success rates often mask a fundamental reliance on a subset of modalities. This establishes a principled foundation for mechanistic analysis and future defense development.

cs.LG↗
arXiv2026

SEAnet: A Deep Learning Architecture for Data Series Similarity Search

A key operation for massive data series collection analysis is similarity search. According to recent studies, SAX-based indexes offer state-of-the-art performance for similarity search tasks. However, their performance lags under high-frequency, weakly correlated, excessively noisy, or other dataset-specific properties. In this work, we propose Deep Embedding Approximation (DEA), a novel family of data series summarization techniques based on deep neural networks. Moreover, we describe SEAnet, a novel architecture especially designed for learning DEA, that introduces the Sum of Squares preservation property into the deep network design. We further enhance SEAnet with SEAtrans encoder. Finally, we propose novel sampling strategies, SEAsam and SEAsamE, that allow SEAnet to effectively train on massive datasets. Comprehensive experiments on 7 diverse synthetic and real datasets verify the advantages of DEA learned using SEAnet in providing high-quality data series summarizations and similarity search results.

cs.DB↗
arXiv2026

Decoder-based Sense Knowledge Distillation

Large language models (LLMs) learn contextual embeddings that capture rich semantic information, yet they often overlook structured lexical knowledge such as word senses and relationships. Prior work has shown that incorporating sense dictionaries can improve knowledge distillation for encoder models, but their application to decoder as generative models remains challenging. In this paper, we introduce Decoder-based Sense Knowledge Distillation (DSKD), a framework that integrates lexical resources into the training of decoder-style LLMs without requiring dictionary lookup at inference time. Extensive experiments on diverse benchmarks demonstrate that DSKD significantly enhances knowledge distillation performance for decoders, enabling generative models to inherit structured semantics while maintaining efficient training.

cs.CL↗
arXiv2026

SEMixer: Semantics Enhanced MLP-Mixer for Multiscale Mixing and Long-term Time Series Forecasting

Modeling multiscale patterns is crucial for long-term time series forecasting (TSF). However, redundancy and noise in time series, together with semantic gaps between non-adjacent scales, make the efficient alignment and integration of multi-scale temporal dependencies challenging. To address this, we propose SEMixer, a lightweight multiscale model designed for long-term TSF. SEMixer features two key components: a Random Attention Mechanism (RAM) and a Multiscale Progressive Mixing Chain (MPMC). RAM captures diverse time-patch interactions during training and aggregates them via dropout ensemble at inference, enhancing patch-level semantics and enabling MLP-Mixer to better model multi-scale dependencies. MPMC further stacks RAM and MLP-Mixer in a memory-efficient manner, achieving more effective temporal mixing. It addresses semantic gaps across scales and facilitates better multiscale modeling and forecasting performance. We not only validate the effectiveness of SEMixer on 10 public datasets, but also on the \textit{2025 CCF AlOps Challenge} based on 21GB real wireless network data, where SEMixer achieves third place. The code is available at the link https://github.com/Meteor-Stars/SEMixer.

cs.LG↗
arXiv2026

ParisKV: Fast and Drift-Robust KV-Cache Retrieval for Long-Context LLMs

KV-cache retrieval is essential for long-context LLM inference, yet existing methods struggle with distribution drift and high latency at scale. We introduce ParisKV, a drift-robust, GPU-native KV-cache retrieval framework based on collision-based candidate selection, followed by a quantized inner-product reranking estimator. For million-token contexts, ParisKV supports CPU-offloaded KV caches via Unified Virtual Addressing (UVA), enabling on-demand top-$k$ fetching with minimal overhead. ParisKV matches or outperforms full attention quality on long-input and long-generation benchmarks. It achieves state-of-the-art long-context decoding efficiency: it matches or exceeds full attention speed even at batch size 1 for long contexts, delivers up to 2.8$\times$ higher throughput within full attention's runnable range, and scales to million-token contexts where full attention runs out of memory. At million-token scale, ParisKV reduces decode latency by 17$\times$ and 44$\times$ compared to MagicPIG and PQCache, respectively, two state-of-the-art KV-cache Top-$k$ retrieval baselines, code is available at https://github.com/amy-77/ParisKV/tree/main.

cs.LG↗
arXiv2026

A Lightweight Sparse Interaction Network for Time Series Forecasting

Recent work shows that linear models can outperform several transformer models in long-term time-series forecasting (TSF). However, instead of explicitly performing temporal interaction through self-attention, linear models implicitly perform it based on stacked MLP structures, which may be insufficient in capturing the complex temporal dependencies and their performance still has potential for improvement. To this end, we propose a Lightweight Sparse Interaction Network (LSINet) for TSF task. Inspired by the sparsity of self-attention, we propose a Multihead Sparse Interaction Mechanism (MSIM). Different from self-attention, MSIM learns the important connections between time steps through sparsity-induced Bernoulli distribution to capture temporal dependencies for TSF. The sparsity is ensured by the proposed self-adaptive regularization loss. Moreover, we observe the shareability of temporal interactions and propose to perform Shared Interaction Learning (SIL) for MSIM to further enhance efficiency and improve convergence. LSINet is a linear model comprising only MLP structures with low overhead and equipped with explicit temporal interaction mechanisms. Extensive experiments on public datasets show that LSINet achieves both higher accuracy and better efficiency than advanced linear models and transformer models in TSF tasks. The code is available at the link https://github.com/Meteor-Stars/LSINet.

cs.LG↗
arXiv2025

Multi-period Learning for Financial Time Series Forecasting

Time series forecasting is important in finance domain. Financial time series (TS) patterns are influenced by both short-term public opinions and medium-/long-term policy and market trends. Hence, processing multi-period inputs becomes crucial for accurate financial time series forecasting (TSF). However, current TSF models either use only single-period input, or lack customized designs for addressing multi-period characteristics. In this paper, we propose a Multi-period Learning Framework (MLF) to enhance financial TSF performance. MLF considers both TSF's accuracy and efficiency requirements. Specifically, we design three new modules to better integrate the multi-period inputs for improving accuracy: (i) Inter-period Redundancy Filtering (IRF), that removes the information redundancy between periods for accurate self-attention modeling, (ii) Learnable Weighted-average Integration (LWI), that effectively integrates multi-period forecasts, (iii) Multi-period self-Adaptive Patching (MAP), that mitigates the bias towards certain periods by setting the same number of patches across all periods. Furthermore, we propose a Patch Squeeze module to reduce the number of patches in self-attention modeling for maximized efficiency. MLF incorporates multiple inputs with varying lengths (periods) to achieve better accuracy and reduces the costs of selecting input lengths during training. The codes and datasets are available at https://github.com/Meteor-Stars/MLF.

q-fin.ST↗
arXiv2025

Aokana: A GPU-Driven Voxel Rendering Framework for Open World Games

Voxels are among the most popular 3D geometric representations today. Due to their intuitiveness and ease-of-editing, voxels have been widely adopted in stylized games and low-cost independent games. However, the high storage cost of voxels, along with the significant time overhead associated with large-scale voxel rendering, limits the further development of open-world voxel games. In this paper, we introduce Aokana, a GPU-Driven Voxel Rendering Framework for Open World Games. Aokana is based on a Sparse Voxel Directed Acyclic Graph (SVDAG). It incorporates a Level-of-Details (LOD) mechanism and a streaming system, enabling seamless map loading as players traverse the open-world game environment. We also designed a corresponding high-performance GPU-driven voxel rendering pipeline to support real-time rendering of the voxel scenes that contain tens of billions of voxels. Aokana can be directly applied to existing game engines and easily integrated with mesh-based rendering methods, demonstrating its practical applicability in game development. Experimental evaluations show that, with increasing voxel scene resolution, Aokana can reduce memory usage by up to ninefold and achieves rendering speeds up to 4.8 times faster than those of previous state-of-the-art approaches.

cs.GR↗
arXiv2025

Multi-Sense Embeddings for Language Models and Knowledge Distillation

Transformer-based large language models (LLMs) rely on contextual embeddings which generate different (continuous) representations for the same token depending on its surrounding context. Nonetheless, words and tokens typically have a limited number of senses (or meanings). We propose multi-sense embeddings as a drop-in replacement for each token in order to capture the range of their uses in a language. To construct a sense embedding dictionary, we apply a clustering algorithm to embeddings generated by an LLM and consider the cluster centers as representative sense embeddings. In addition, we propose a novel knowledge distillation method that leverages the sense dictionary to learn a smaller student model that mimics the senses from the much larger base LLM model, offering significant space and inference time savings, while maintaining competitive performance. Via thorough experiments on various benchmarks, we showcase the effectiveness of our sense embeddings and knowledge distillation approach. We share our code at https://github.com/Qitong-Wang/SenseDict

cs.CL↗
arXiv2025

Automated Data Quality Validation in an End-to-End GNN Framework

Ensuring data quality is crucial in modern data ecosystems, especially for training or testing datasets in machine learning. Existing validation approaches rely on computing data quality metrics and/or using expert-defined constraints. Although there are automated constraint generation methods, they are often incomplete and may be too strict or too soft, causing false positives or missed errors, thus requiring expert adjustment. These methods may also fail to detect subtle data inconsistencies hidden by complex interdependencies within the data. In this paper, we propose DQuag, an end-to-end data quality validation and repair framework based on an improved Graph Neural Network (GNN) and multi-task learning. The proposed method incorporates a dual-decoder design: one for data quality validation and the other for data repair. Our approach captures complex feature relationships within tabular datasets using a multi-layer GNN architecture to automatically detect explicit and hidden data errors. Unlike previous methods, our model does not require manual input for constraint generation and learns the underlying feature dependencies, enabling it to identify complex hidden errors that traditional systems often miss. Moreover, it can recommend repair values, improving overall data quality. Experimental results validate the effectiveness of our approach in identifying and resolving data quality issues. The paper appeared in EDBT 2025.

cs.DB↗
arXiv2025

LeaFi: Data Series Indexes on Steroids with Learned Filters

The ever-growing collections of data series create a pressing need for efficient similarity search, which serves as the backbone for various analytics pipelines. Recent studies have shown that tree-based series indexes excel in many scenarios. However, we observe a significant waste of effort during search, due to suboptimal pruning. To address this issue, we introduce LeaFi, a novel framework that uses machine learning models to boost pruning effectiveness of tree-based data series indexes. These models act as learned filters, which predict tight node-wise distance lower bounds that are used to make pruning decisions, thus, improving pruning effectiveness. We describe the LeaFi-enhanced index building algorithm, which selects leaf nodes and generates training data to insert and train machine learning models, as well as the LeaFi-enhanced search algorithm, which calibrates learned filters at query time to support the user-defined quality target of each query. Our experimental evaluation, using two different tree-based series indexes and five diverse datasets, demonstrates the advantages of the proposed approach. LeaFi-enhanced data-series indexes improve pruning ratio by up to 20x and search time by up to 32x, while maintaining a target recall of 99%.

cs.DB↗
arXiv2024

Beyond Accuracy: On the Effects of Fine-tuning Towards Vision-Language Model's Prediction Rationality

Vision-Language Models (VLMs), such as CLIP, have already seen widespread applications. Researchers actively engage in further fine-tuning VLMs in safety-critical domains. In these domains, prediction rationality is crucial: the prediction should be correct and based on valid evidence. Yet, for VLMs, the impact of fine-tuning on prediction rationality is seldomly investigated. To study this problem, we proposed two new metrics called Prediction Trustworthiness and Inference Reliability. We conducted extensive experiments on various settings and observed some interesting phenomena. On the one hand, we found that the well-adopted fine-tuning methods led to more correct predictions based on invalid evidence. This potentially undermines the trustworthiness of correct predictions from fine-tuned VLMs. On the other hand, having identified valid evidence of target objects, fine-tuned VLMs were more likely to make correct predictions. Moreover, the findings are also consistent under distributional shifts and across various experimental settings. We hope our research offer fresh insights to VLM fine-tuning.

cs.LG↗
arXiv2024

DumpyOS: A Data-Adaptive Multi-ary Index for Scalable Data Series Similarity Search

Data series indexes are necessary for managing and analyzing the increasing amounts of data series collections that are nowadays available. These indexes support both exact and approximate similarity search, with approximate search providing high-quality results within milliseconds, which makes it very attractive for certain modern applications. Reducing the pre-processing (i.e., index building) time and improving the accuracy of search results are two major challenges. DSTree and the iSAX index family are state-of-the-art solutions for this problem. However, DSTree suffers from long index building times, while iSAX suffers from low search accuracy. In this paper, we identify two problems of the iSAX index family that adversely affect the overall performance. First, we observe the presence of a proximity-compactness trade-off related to the index structure design (i.e., the node fanout degree), significantly limiting the efficiency and accuracy of the resulting index. Second, a skewed data distribution will negatively affect the performance of iSAX. To overcome these problems, we propose Dumpy, an index that employs a novel multi-ary data structure with an adaptive node splitting algorithm and an efficient building workflow. Furthermore, we devise Dumpy-Fuzzy as a variant of Dumpy which further improves search accuracy by proper duplication of series. To fully leverage the potential of modern hardware including multicore CPUs and Solid State Drives (SSDs), we parallelize Dumpy to DumpyOS with sophisticated indexing and pruning-based querying algorithms. An optimized approximate search algorithm, DumpyOS-F which prominently improves the search accuracy without violating the index, is also proposed.

cs.DB↗
arXiv2024

Enhancing Adaptive Deep Networks for Image Classification via Uncertainty-aware Decision Fusion

Handling varying computational resources is a critical issue in modern AI applications. Adaptive deep networks, featuring the dynamic employment of multiple classifier heads among different layers, have been proposed to address classification tasks under varying computing resources. Existing approaches typically utilize the last classifier supported by the available resources for inference, as they believe that the last classifier always performs better across all classes. However, our findings indicate that earlier classifier heads can outperform the last head for certain classes. Based on this observation, we introduce the Collaborative Decision Making (CDM) module, which fuses the multiple classifier heads to enhance the inference performance of adaptive deep networks. CDM incorporates an uncertainty-aware fusion method based on evidential deep learning (EDL), that utilizes the reliability (uncertainty values) from the first c-1 classifiers to improve the c-th classifier' accuracy. We also design a balance term that reduces fusion saturation and unfairness issues caused by EDL constraints to improve the fusion quality of CDM. Finally, a regularized training strategy that uses the last classifier to guide the learning process of early classifiers is proposed to further enhance the CDM module's effect, called the Guided Collaborative Decision Making (GCDM) framework. The experimental evaluation demonstrates the effectiveness of our approaches. Results on ImageNet datasets show CDM and GCDM obtain 0.4% to 2.8% accuracy improvement (under varying computing resources) on popular adaptive networks. The code is available at the link https://github.com/Meteor-Stars/GCDM_AdaptiveNet.

cs.CV↗