arXiv subjects

Yanxuan Yu

Publications and source records attributed to Yanxuan Yu.

At least 19 recordsLinked to original sources

arXiv2026

Privacy-Preserving Topology-Guided Safety for LLM-Based Multi-Agent Systems via Federated Graph Learning

Topology-guided safeguards for LLM-based multi-agent systems (MAS) train a GNN over the inter-agent communication graph to localize risky agents and intervene on the topology---but they assume one operator can pool all labeled traces. Across organizations that assumption breaks: episodes contain private prompts, tool outputs, and proprietary workflows, and no silo alone sees the full attack distribution. We cast privacy-preserving MAS safeguarding as graph federated learning and instantiate FGLGuard: each operator fits an edge-featured graph attention detector on its own judge-labeled episode graphs and shares only model updates. The method couples a proximal local objective for non-IID clients, domain-balanced aggregation, over-refusal-constrained threshold calibration, corroborated upstream scoring, and a guarded rewrite for blocked answers. Federation is not optional: off-the-shelf transfer collapses under distribution shift (AUROC 0.51 to 0.70 only after in-domain retraining), so a deployable guard must adapt on each site's private traces. On Agent-SafetyBench, R-Judge, and AgentDojo, federated FGLGuard exceeds the in-domain centralized ceiling on all three benchmarks without pooling any data---where unsupervised anomaly guards and local-only training fail. One guard federated across four different-domain operators comes within 0.03 AUROC of multi-domain centralization, while any single-domain guard collapses on the others. Live FGLGuard cuts AgentDojo's ground-truth attack-success rate by 43% at near-unguarded utility, zero API cost, and negligible capability loss.

cs.CR↗

arXiv2026

HSMLA: Hierarchical Softmax Multi-scale Linear Attention for Efficient Vision Transformers

Vision transformers face significant computational overheads in high-resolution dense prediction due to the quadratic complexity of self-attention. Linear attention offers efficiency but sacrifices local context modeling. We propose \textbf{HSMLA (Hierarchical Softmax Multi-scale Linear Attention)}, which combines ReLU-based linear attention for global context, selective softmax refinement for critical local features, and multi-scale token representations via depthwise convolutions. HSMLA achieves superior accuracy-efficiency trade-offs: up to $4.2\times$ inference-time speedup across dense prediction tasks, $87.3%$ Dice with $3.2\times$ speedup on CT organ segmentation, and $94.2%$ AUC with $4.1\times$ speedup on pathology WSI.

cs.CV↗

arXiv2026

RUBRIC: Realism--Utility Balanced Ranking for Imbalanced Classification

Class imbalance poses a fundamental challenge in risk-sensitive applications such as fraud detection and medical diagnosis, where minority-class samples are scarce yet critical for accurate classification. Existing oversampling methods generate synthetic samples to rebalance class distributions; however, they often produce large numbers of low-quality candidates that distort decision boundaries or introduce artifacts, leading to overfitting and degraded generalization. In this work, we introduce RUBRIC, a generator-agnostic filtering framework that formulates synthetic sample selection as a quality-over-quantity optimization problem. RUBRIC ranks candidates using a realism-utility trade-off: realism is quantified by a learned discriminator that distinguishes real samples from synthetic samples, while utility captures proximity to the decision boundary through a concave margin-based scoring function. We show that, under mild regularity conditions, the proposed filtering strategy monotonically tightens the generalization bound for margin-based classifiers by jointly reducing distribution shift and suppressing near-negative tail contributions. Through extensive experiments on credit-card fraud detection and other imbalanced benchmarks, we demonstrate that RUBRIC improves F1-macro and recall while maintaining comparable ROC-AUC across several generators. We also provide explicit lambda-sensitivity analysis to show how users can recover AUPRC when ranking quality is prioritized.

cs.LG↗

arXiv2026

OLIVE: Online Low-Rank Incremental Learning for Efficient Adaptive Exoskeletons

Wearable exoskeleton systems hold promise for restoring mobility in individuals with physical impairments, yet most existing controllers rely on static gait policies that cannot adapt to dynamic real-world environments or individual user characteristics. We present OLIVE (Online Low-rank Incremental Learning for Efficient Adaptive Exoskeletons), a parameter-efficient online adaptation framework that continuously personalizes exoskeleton control during deployment. OLIVE decomposes the adaptive component of the control policy into a low-rank residual form $\Delta W = A_t B_t^\top$ with rank $r \ll \min(d,k)$, reducing the online update cost from $\mathcal{O}(dk)$ to $\mathcal{O}(r(d+k))$ while preserving the stability of a pretrained base controller $W_0$. Parameters are updated through a reward-shaped policy gradient driven purely by on-body sensor feedback, including EMG, IMU, and vibration signals, eliminating dependence on offline reference trajectories. A gating mechanism modulates the strength of personalization based on the contextual state, while a dynamic rank scheduler adapts the update dimensionality to terrain complexity. It allocates minimal capacity on simple flat terrain and expands to higher-rank updates on demanding uneven surfaces, enabling robust performance across flat walking, stair navigation, slopes, and uneven terrain. Experiments on the wearable platform demonstrate that OLIVE achieves improvements of 13, 22, and 15 percentage points in gait smoothness, effort reduction, and motion stability over the strongest baseline, respectively. It converges within approximately 1,800 walking steps with an end-to-end latency of 7.4 ms. Our code implementation is available at https://github.com/FastLM/OLIVE.

cs.RO↗

arXiv2026

Thoughts-as-Planning: Latent World Models for Chain-of-Thoughts Optimization via Reinforcement Planning

The success of large language models (LLMs) across diverse NLP tasks has elevated the importance of reasoning chain optimization as a critical step in aligning model behavior with task objectives. Existing reasoning chain tuning methods often rely on black-box heuristics or gradient-free search, which lack interpretability, generalization, and sample efficiency. In this work, we introduce \textbf{Thoughts-as-Planning}, a novel framework that formalizes reasoning chain optimization as a sequential decision-making process over a latent semantic space. We model the LLM as a partially observable environment and learn a latent world model that simulates the effect of reasoning chain edits on downstream outputs. A proximity-preserving embedding space is constructed to encode reasoning chain-response dynamics, enabling planning via gradient descent or reinforcement learning. Our method supports multi-scale abstraction, allowing reasoning chain edits at token, segment, and instruction levels to be integrated into a unified planner. Through extensive experiments on language understanding and generation tasks, we demonstrate that Thoughts-as-Planning outperforms state-of-the-art reasoning chain tuning baselines in efficiency, robustness, and generalization, while offering interpretability through its structured planning trajectory. Our code is available at https://github.com/FastLM/Thoughts-as-Planning.

cs.CL↗

arXiv2026

Accelerating Frequency Domain Diffusion Models with Error-Feedback Event-Driven Caching

Diffusion models achieve remarkable success in time series generation. However, slow inference limits their practical deployment. We propose E$^2$-CRF (Error-Feedback Event-Driven Cumulative Residual Feature caching) to accelerate frequency domain diffusion models. Our method exploits two structural properties: (1) spectral localization, where signal energy concentrates in low frequencies, and (2) mirror symmetry, which halves the effective frequency dimension. E$^2$-CRF uses a closed-loop error-feedback system that adaptively caches transformer KV features across diffusion steps. We trigger recomputation using event-driven residual dynamics instead of fixed schedules. Our method selectively recomputes high-energy or rapidly-changing tokens while reusing cached features for stable high-frequency components. E$^2$-CRF achieves ~2.2 speedup while maintaining sample quality. We demonstrate effectiveness on 5 datasets. Our caching strategy naturally aligns with the diffusion process's structure-to-detail progression. We include sufficient-condition error and complexity bounds under standard regularity assumptions (Appendix), alongside empirical validation. Our code is available at https://github.com/NoakLiu/FastFourierDiffusion and is also integrated in https://github.com/NoakLiu/FastCache-xDiT.

cs.LG↗

arXiv2026

MKA: Memory-Keyed Attention for Efficient Long-Context Reasoning

As long-context language modeling becomes increasingly important, the cost of maintaining and attending to large Key/Value (KV) caches grows rapidly, becoming a major bottleneck in both training and inference. While prior works such as Multi-Query Attention (MQA) and Multi-Latent Attention (MLA) reduce memory by sharing or compressing KV features, they often trade off representation quality or incur runtime overhead. We propose Memory-Keyed Attention (MKA), a hierarchical attention mechanism that integrates multi-level KV caches (local, session, and long-term) and learns to route attention across them dynamically. We further introduce Route-Fused MKA (FastMKA), a broadcast-routed variant that fuses memory sources before attention computation for improved efficiency. Experiments on different sequence lengths show that FastMKA achieves a favorable accuracy-efficiency trade-off: comparable perplexity to MLA while achieving up to 5x faster training throughput and 1.8x lower evaluation latency. These results highlight MKA as a practical and extensible framework for efficient long-context attention.

cs.LG↗

arXiv2026

AdaCorrection: Adaptive Offset Cache Correction for Accurate Diffusion Transformers

Diffusion Transformers (DiTs) achieve state-of-the-art performance in high-fidelity image and video generation but suffer from expensive inference due to their iterative denoising structure. While prior methods accelerate sampling by caching intermediate features, they rely on static reuse schedules or coarse-grained heuristics, which often lead to temporal drift and cache misalignment that significantly degrade generation quality. We introduce \textbf{AdaCorrection}, an adaptive offset cache correction framework that maintains high generation fidelity while enabling efficient cache reuse across Transformer layers during diffusion inference. At each timestep, AdaCorrection estimates cache validity with lightweight spatio-temporal signals and adaptively blends cached and fresh activations. This correction is computed on-the-fly without additional supervision or retraining. Our approach achieves strong generation quality with minimal computational overhead, maintaining near-original FID while providing moderate acceleration. Experiments on image and video diffusion benchmarks show that AdaCorrection consistently improves generation performance. Our code has been integrated into FastCache-xDiT at https://github.com/NoakLiu/FastCache-xDiT.

cs.CV↗

arXiv2025

CXL-SpecKV: A Disaggregated FPGA Speculative KV-Cache for Datacenter LLM Serving

Large Language Models (LLMs) have revolutionized natural language processing tasks, but their deployment in datacenter environments faces significant challenges due to the massive memory requirements of key-value (KV) caches. During the autoregressive decoding process, KV caches consume substantial GPU memory, limiting batch sizes and overall system throughput. To address these challenges, we propose \textbf{CXL-SpecKV}, a novel disaggregated KV-cache architecture that leverages Compute Express Link (CXL) interconnects and FPGA accelerators to enable efficient speculative execution and memory disaggregation. Our approach introduces three key innovations: (i) a CXL-based memory disaggregation framework that offloads KV-caches to remote FPGA memory with low latency, (ii) a speculative KV-cache prefetching mechanism that predicts and preloads future tokens' cache entries, and (iii) an FPGA-accelerated KV-cache compression and decompression engine that reduces memory bandwidth requirements by up to 4$\times$. When evaluated on state-of-the-art LLM models, CXL-SpecKV achieves up to 3.2$\times$ higher throughput compared to GPU-only baselines, while reducing memory costs by 2.8$\times$ and maintaining accuracy. Our system demonstrates that intelligent memory disaggregation combined with speculative execution can effectively address the memory wall challenge in large-scale LLM serving. Our code implementation has been open-sourced at https://github.com/FastLM/CXL-SpecKV.

cs.AI↗

arXiv2025

Boundary-Aware Adversarial Filtering for Reliable Diagnosis under Extreme Class Imbalance

We study classification under extreme class imbalance where recall and calibration are both critical, for example in medical diagnosis scenarios. We propose AF-SMOTE, a mathematically motivated augmentation framework that first synthesizes minority points and then filters them by an adversarial discriminator and a boundary utility model. We prove that, under mild assumptions on the decision boundary smoothness and class-conditional densities, our filtering step monotonically improves a surrogate of F_beta (for beta >= 1) while not inflating Brier score. On MIMIC-IV proxy label prediction and canonical fraud detection benchmarks, AF-SMOTE attains higher recall and average precision than strong oversampling baselines (SMOTE, ADASYN, Borderline-SMOTE, SVM-SMOTE), and yields the best calibration. We further validate these gains across multiple additional datasets beyond MIMIC-IV. Our successful application of AF-SMOTE to a healthcare dataset using a proxy label demonstrates in a disease-agnostic way its practical value in clinical situations, where missing true positive cases in rare diseases can have severe consequences.

cs.LG↗

arXiv2025

CSV-Decode: Certifiable Sub-Vocabulary Decoding for Efficient Large Language Model Inference

Large language models face significant computational bottlenecks during inference due to the expensive output layer computation over large vocabularies. We present CSV-Decode, a novel approach that uses geometric upper bounds to construct small sub-vocabularies for each decoding step, enabling efficient sparse computation while maintaining dual correctness guarantees: exact top-$k$ certification and $\varepsilon$-certified softmax approximations. Our method clusters vocabulary embeddings offline and uses centroid-plus-radius bounds to identify which tokens can be safely omitted from computation. We provide a complete system implementation with sparse GEMV kernels, multi-GPU sharding, and CUDA Graph optimization. Experimental results demonstrate significant speedup over full vocabulary decoding while maintaining distributional guarantees and low fallback rates. Our code implementation available at \href{https://github.com/FastLM/CSV-Decode}{https://github.com/FastLM/CSV-Decode}.

cs.CL↗

arXiv2025

$\pi$-Attention: Online Efficient Sparse Transformers for Long-Context Modeling

Sparse attention is crucial in long-context Transformers, which restricts each token to a limited neighborhood and thereby reduces the quadratic cost of full self-attention. Local windows capture nearby context effectively, yet they induce a receptive-field bottleneck for dependencies beyond the window, limiting long-range modeling under moderate depth. In this paper, we propose $\pi$-Attention, an \emph{online efficient} sparse attention operator: as tokens arrive, each step maintains a streaming working set of local neighbors plus a $\pi$-indexed long-range fetch, fused by an adaptive prior under a shared softmax. Rather than materializing a global sparse mask in advance, $\pi$-Attention computes attention on the live working set with hierarchy-aware IO. We analyze causal reachability and minimum depth under this online rule, and show per-step cost remains $\mathcal{O}(k)$. Experiments on language modeling, Long Range Arena, and efficiency profiling---across 4K--32K context lengths---show consistent gains over local-window and other sparse baselines, approaching dense attention quality at linear cost.

cs.CL↗

arXiv2025

SPI: Query-Depth-Adaptive Indexing for Streaming RAG in Vector Databases

Vector databases (VecDBs) are increasingly deployed in retrieval-augmented generation (RAG) pipelines where query processing and document ingestion occur concurrently. The index layer needs to provide low-latency search while incorporating new vectors without frequent global rebuilding. Existing VecDB pipelines typically operate within a uniform representation regime, despite substantial variation in the semantic granularity required across queries. This motivates an index design that supports incremental updates while adapting retrieval depth to query distribution and complexity. We propose \textbf{Semantic Pyramid Indexing (SPI)}, a VecDB-layer indexing framework that organizes embeddings into $L$ semantically aligned resolution levels and selects retrieval depth per query via a lightweight uncertainty-aware controller. SPI supports progressive coarse-to-fine ANN search, level-wise streaming insertion without global rebuilds, and distributed execution through LSH partitioning with asynchronous gRPC coordination. Unlike hierarchical ANN structures with fixed traversal rules (e.g., SPANN), SPI adapts resolution at query time while remaining compatible with FAISS and Qdrant backends. On MS MARCO and Natural Questions, SPI achieves competitive Recall@10 with lower latency under the same dense encoder family, yielding a \textbf{1.4--2.3$\times$} average retrieval latency reduction under fixed Recall@10 targets relative to comparable approximate-ANN baselines. A prototype scaling study up to 8 nodes shows $6.2\times$ throughput scaling (${\approx}73\%$ efficiency); the 16-node configuration is included for completeness but shows diminishing efficiency. We provide a top-$K$ stability guarantee: queries with sufficient retrieval margin return an identical top-$K$ set at a shallower level. Code and configurations are available at https://github.com/FastLM/SPI_VecDB.

cs.CL↗

arXiv2025

Cognitive Load Traces as Symbolic and Visual Accounts of Deep Model Cognition

We propose \textbf{Cognitive Load Traces} (CLTs) as a mid-level interpretability framework for deep models, inspired by Cognitive Load Theory in human cognition. CLTs are defined as symbolic, temporally varying functions that quantify model-internal resource allocation. Formally, we represent CLTs as a three-component stochastic process $(\mathrm{IL}_t, \mathrm{EL}_t, \mathrm{GL}_t)$, corresponding to \emph{Intrinsic}, \emph{Extraneous}, and \emph{Germane} load. Each component is instantiated through measurable proxies such as attention entropy, KV-cache miss ratio, representation dispersion, and decoding stability. We propose both symbolic formulations and visualization methods (load curves, simplex diagrams) that enable interpretable analysis of reasoning dynamics. Experiments on reasoning and planning benchmarks show that CLTs predict error-onset, reveal cognitive strategies, and enable load-guided interventions that improve reasoning efficiency by 15-30\% while maintaining accuracy.

cs.AI↗

arXiv2025

HSGM: Hierarchical Segment-Graph Memory for Scalable Long-Text Semantics

Semantic parsing of long documents remains challenging due to quadratic growth in pairwise composition and memory requirements. We introduce \textbf{Hierarchical Segment-Graph Memory (HSGM)}, a novel framework that decomposes an input of length $N$ into $M$ meaningful segments, constructs \emph{Local Semantic Graphs} on each segment, and extracts compact \emph{summary nodes} to form a \emph{Global Graph Memory}. HSGM supports \emph{incremental updates} -- only newly arrived segments incur local graph construction and summary-node integration -- while \emph{Hierarchical Query Processing} locates relevant segments via top-$K$ retrieval over summary nodes and then performs fine-grained reasoning within their local graphs. Theoretically, HSGM reduces worst-case complexity from $O(N^2)$ to $O\!\left(N\,k + (N/k)^2\right)$, with segment size $k \ll N$, and we derive Frobenius-norm bounds on the approximation error introduced by node summarization and sparsification thresholds. Empirically, on three benchmarks -- long-document AMR parsing, segment-level semantic role labeling (OntoNotes), and legal event extraction -- HSGM achieves \emph{2--4$\times$ inference speedup}, \emph{$>60\%$ reduction} in peak memory, and \emph{$\ge 95\%$} of baseline accuracy. Our approach unlocks scalable, accurate semantic modeling for ultra-long texts, enabling real-time and resource-constrained NLP applications.

cs.AI↗

arXiv2025

TinyServe: Query-Aware Cache Selection for Efficient LLM Serving

Serving large language models (LLMs) efficiently remains challenging due to the high memory and latency overhead of key-value (KV) cache access during autoregressive decoding. We present \textbf{TinyServe}, a lightweight and extensible serving system for deploying tiny LLMs (e.g., TinyLLaMA, GPT2-345M) with support for structured KV sparsity, plugin-based token selection, and hardware-efficient attention kernels. Unlike prior simulation frameworks, TinyServe executes real-time decoding with configurable sparsity strategies and fine-grained instrumentation. To reduce decoding cost, we introduce a \textit{query-aware page selection} mechanism that leverages bounding-box metadata to estimate attention relevance between the query and KV cache blocks. This enables selective KV loading with minimal overhead and no model modifications. Our fused CUDA kernel integrates page scoring, sparse memory access, and masked attention in a single pass. Experiments show that TinyServe achieves up to \textbf{3.4x} speedup and over \textbf{2x} memory savings with negligible accuracy drop. Additional analysis of cache reuse, page hit rate, and multi-GPU scaling confirms its practicality as an efficient system-level design for LLM training and inference research on resource-constrained hardware.

cs.DC↗

arXiv2025

SemToken: Semantic-Aware Tokenization for Efficient Long-Context Language Modeling

Tokenization plays a critical role in language modeling, yet existing approaches such as Byte-Pair Encoding (BPE) or WordPiece operate purely on frequency statistics, ignoring the underlying semantic structure of text. This leads to over-tokenization of semantically redundant spans and underutilization of contextual coherence, particularly in long-context scenarios. In this work, we propose \textbf{SemToken}, a semantic-aware tokenization framework that jointly reduces token redundancy and improves computation efficiency. SemToken first extracts contextual semantic embeddings via lightweight encoders and performs local semantic clustering to merge semantically equivalent tokens. Then, it allocates heterogeneous token granularity based on semantic density, allowing finer-grained tokenization in content-rich regions and coarser compression in repetitive or low-entropy spans. SemToken can be seamlessly integrated with modern language models and attention acceleration methods. Experiments on long-context language modeling benchmarks such as WikiText-103 and LongBench show that SemToken achieves up to $2.4\times$ reduction in token count and $1.9\times$ speedup, with negligible or no degradation in perplexity and downstream accuracy. Our findings suggest that semantic structure offers a promising new axis for optimizing tokenization and computation in large language models.

cs.CL↗

arXiv2025

QuickMerge++: Fast Token Merging with Autoregressive Prior

As generative models scale to larger inputs across language, vision, and video domains, the cost of token-level computation has become a key bottleneck. While prior work suggests that only a subset of tokens significantly influence downstream predictions, most token selection methods are static, modality-specific, or incompatible with autoregressive generation. In this paper, we propose QuickMerge, a lightweight token merging framework designed for efficient next-token prediction. QuickMerge dynamically selects a reduced number of tokens based on attention norm magnitude, guided by an entropy-based budget estimator. To preserve autoregressive compatibility, we introduce a lightweight transformer prior trained over the merged token sequence. By combining semantic salience estimation, flexible token budgets, and AR alignment, QuickMerge enables accurate generation with fewer tokens. We evaluate QuickMerge across multi-modality domains, demonstrating consistent improvements in compute-accuracy tradeoffs. Specifically, QuickMerge reduces token counts sustantially while matching as well as exceeding the performance of learned tokenizers and fixed-patch baselines.

cs.AI↗