arXiv subjects

Yuchen Li

Publications and source records attributed to Yuchen Li.

At least 19 recordsLinked to original sources

arXiv2026

InfiniLoRA: Disaggregated Multi-LoRA Serving for Large Language Models

LoRA enables efficient customization of LLMs and is widely used in multi-tenant and multi-task serving. However, emerging model architectures such as MoE significantly increase LoRA memory cost, making existing coupled LoRA serving designs poorly scalable and prone to tail-latency inflation. We present InfiniLoRA, a disaggregated LoRA serving system that decouples LoRA execution from base-model inference. InfiniLoRA introduces a shared LoRA Server with parallelism-aware execution, SLO-driven provisioning, and critical-path optimizations, including GPU-initiated communication and hardware-specialized LoRA kernels. Experiments show that InfiniLoRA can achieve an average $3.05\times$ increase in serviceable request rate under strict latency SLOs, and improve the percentage of LoRA adapters satisfying the SLO requirement by 54.0\%.

cs.DC↗

arXiv2026

Why Does Train-Validation Separation Emerge? Update-Pressure Density Dynamics in Pretrained Backbones

During fine-tuning, training accuracy often keeps improving after validation accuracy levels off, widening the train-validation gap. We propose a support-transfer account: early updates reuse structure shared by many examples, whereas later fitting relies on narrower structure whose benefits transfer less to held-out data. From fixed training examples, we normalize per-example gradient magnitudes over selected parameters and measure allocation dispersion across examples ($D_{\mathrm{den}}$) and within labels, weighted by class sample counts ($D_{\mathrm{within}}^{\mathrm{wt}}$). A local model gives sufficient conditions for dispersion and the cross-entropy gap to grow together, separating uniform score scaling; additional decision-boundary conditions characterize accuracy-gap growth. In a five-level hierarchy, increasing the share of private features, unique to individual examples, from 0.3 to 0.5 to 0.7 raises the final accuracy gap from 0.185 to 0.331 to 0.527 and increases unnormalized dispersion increments. Tests retaining one feature level at a time show a widening validation-minus-training difference in the private level's loss share. After the validation-loss minimum, the cross-entropy gap continues to grow without a decline in mean validation accuracy. Across 90 runs of RoBERTa, DeBERTa, and Qwen3-1.7B on six language tasks, both readouts correlate positively with the accuracy gap in every run; raw correlations of one-epoch changes are positive for $D_{\mathrm{den}}$ in 88/90 runs. Both readouts also retain positive associations in 15 ResNet-18 runs across three vision datasets. These results support a connection between evolving gradient allocation and train-validation separation.

cs.LG↗

arXiv2026

Where Is the "Self" of a Large Language Model? -- From Self-Index to Self-Reference

Whether a large language model has a self is usually argued without a way to measure it. We turn the question into the measurement of a functional self with two levels familiar from biology: a self-index, which marks material as the model's own and shapes how it is processed, much as nearly all animals tag self-generated signals with a corollary discharge, and self-reference, the ability to recognize what is one's own, as in mirror self-recognition. Each level has criteria that can be tested in any open-weight model. We place the same text in an Assistant, a User, or a System message of the model's own chat template. In Qwen3.8-27B, the representation singles out the model's own message from both external sources, as predicted in advance for early layers, and does so at every one of its 64 layers and in every held-out text, while User and System stay close to each other; asked who wrote a message, the model claims its own (0.77) and assigns both external sources to the user. In three models and four readouts (Qwen3.8-27B without and with thinking, gpt-oss-20b, and gpt-oss-120b), replacing eight coordinates per layer of the model's own message by those of the same text as a User message removes the claim (to at most 0.05), installing them into a User message creates it, and random subspaces of the same dimension do neither; the coordinates are read mainly at the end of the message, before the last third of the layers, and editing them leaves answers about the content of the message and an unrelated multiplication unchanged. Across eighteen open-weight models from nine developers, the two levels are nested: no model recognizes its own messages without a self-index, and neither level follows model size or published benchmark scores. Together, these results support the account that self-reference is a readout of a self-index.

cs.CL↗

arXiv2026

Commutators of a class of $Ψ$-bounded oscillation operators on abstract weighted Orlicz-Morrey spaces endowed with ball-basis and applications

In our previous work \cite{SZ2026}, we established a class of abstract Orlicz-Morrey spaces endowed with a ball-basis and introduced the notion of $Ψ$-bounded oscillation operators. The present paper further develops this theory by investigating sparse domination and weighted estimates for such operators and their commutators within the framework of abstract Orlicz-Morrey spaces endowed with a ball-basis. We obtain pointwise sparse domination results for the operators and establish their boundedness on weighted Orlicz-Morrey spaces. The hypotheses of these conclusions differ from those in the classical Orlicz-Morrey setting and do not rely on dyadic decompositions of Euclidean structure. Moreover, we study the duality of these spaces and prove that the dual of the associated block space is exactly the Orlicz-Morrey space under consideration.

math.FA↗

arXiv2026

No Concept Escapes the Audit: Auditing-Aware Unlearning for Verifiable Concept Erasure in Diffusion Models

Text-to-image diffusion models can generate prohibited content, which motivates concept erasure through machine unlearning. Most erasure methods intervene at the text interface, through prompt modification or localized updates to text-conditioning weights, and they are evaluated by what the model outputs for given prompts. Such evaluation cannot see what the network still encodes. Latent-space auditing, which bypasses text conditioning and probes the denoising network directly, shows that erased concepts remain recoverable from internal representations. We find that this also holds for methods built to be robust against adversarial prompts, and that the problem grows with the number of erased concepts. We propose Auditing-Aware Unlearning for Verifiable Concept Erasure in Diffusion Models (AVCE), a framework that grounds erasure in the model's latent representations. AVCE audits the embedding neighborhood of each concept and condenses the discovered vulnerable directions into an anchor at the weakest geometric point. It edits cross-attention and self-attention projections in closed form at this anchor, then fine-tunes the two pathways with pathway-level auditing losses, using orthogonal gradient projection to consolidate multiple concepts. Experiments on SD v1.5, SDXL, and Flux 1.0 across object, explicit-content, and artistic-style unlearning show that AVCE reduces attack success rates by 5.07x and improves auditing scores by 3.84x over the strongest baseline, while preserving competitive generation quality.

cs.LG↗

arXiv2026

Barrier-Constrained Guidance For Real-time Asynchronous VLA Execution

While asynchronous execution enables VLA models to operate in real-time environments by predicting the next chunk during execution, it inevitably introduces boundary misalignment with the committed action stream. Existing delay-aware solutions address this at the prohibitive cost of extensive retraining, architectural modifications, or fixed-prefix inpainting that inflates computation when low latency is vital. To address these problems, we propose FreeBar, a training-free real-time inference plugin that enables robust asynchronous execution with only marginal additional inference cost by casting chunk alignment in flow-based VLAs as barrier-guided action-space constraint. Specifically, FreeBar introduces a lightweight Control Barrier Function (CBF)-inspired consistency constraint between consecutive action chunks and solves the resulting correction in closed form, avoiding costly backpropagation through the VLA action head. By progressively tightening the consistency bound during generation, FreeBar aligns generated chunks with the committed execution stream while preserving early-stage generative and within-chunk future action flexibility,. Across Kinetix, RoboCasa, and diverse real-world robotic platforms with multiple VLA backbones, FreeBar improves success rates over naive asynchronous execution by 10%--15% on average, with larger gains in highly dynamic scenarios. It also preserves real-time computational efficiency, decreasing inference latency per action chunk over 45% than RTC. Additionally, as a universal plugin, FreeBar remains robust across distinct latency environments, base policies, and hyperparameter settings.

cs.RO↗

arXiv2026

FADE: Frame-Aware Diffusion-Transformer-based Multi-Concept Erasure for Video Unlearning

Text-to-video (T2V) diffusion models can reproduce copyrighted, violent, or explicit content, which motivates concept erasure: removing designated concepts from a pretrained model while preserving its behavior on everything else. Existing T2V erasure methods leave two problems open. Their frame-agnostic suppression can leave isolated frames in which an erased concept resurfaces, a frame-reactivation gap that clip-level averages obscure; and they are usually evaluated with one target concept or category at a time. We propose Frame-Aware Diffusion Erasure (FADE), a multi-concept video unlearning framework. FADE first applies a joint closed-form key/value edit that suppresses all target concepts, then trains per-concept frame-aware low-rank adapters whose strength is gated by the frame index and the denoising timestep to remove residual per-frame leakage. Each adapter is trained with the other targets' prompts as hard negatives, which keeps the concept-specific components of different adapters well separated, and a similarity-based soft router combines the adapters according to the prompt. With 16 concepts (objects, artistic styles, and nudity) erased from a single Wan2.1-T2V-1.3B backbone, FADE reduces the residual accuracy on the object benchmark to 4.9%, against 15.5% for the strongest of eight baselines, while keeping the VBench average within 0.9% of the unedited model. The ranking is unchanged under a VLM judge and a blinded human study, and the advantage over the strongest baseline carries over to prompts that combine several erased concepts, to 30 simultaneously erased celebrity identities, and to Wan2.1-T2V-14B, CogVideoX-2B, and HunyuanVideo-1.5.

cs.CV↗

arXiv2026

InstanceBench: Diagnosing Referential Reasoning and Target Identity in Referring Expression Segmentation

Referring Expression Segmentation (RES) links natural-language descriptions to pixel-level object masks. Yet standard evaluation provides limited insight into instance-level referential reasoning: it does not systematically distinguish referential logics, test target preservation across valid grounding paths, or separate target-selection from mask-generation errors. We introduce InstanceBench, an instance-centered diagnostic benchmark comprising 6,194 images, 9,264 target instances, and 25,077 human-verified expressions. Each target-centric expression set (TCES) fixes the image and target mask while pairing a minimal expression with a same-target variant that uses another valid cue or grounding path. A compact referential-logic taxonomy spans direct target evidence, same-class selection, relational and compositional grounding, and exclusion, while logic-critical construction suppresses simpler shortcuts. Identity-aware metrics measure target retention and set-level success while separating selection from mask-generation errors. Across 22 native-mask RES checkpoints from 18 model families, the strongest checkpoint reaches 67.1% mIoU but only 59.6% All@0.7. Controlled interventions confirm language sensitivity, while failure decomposition identifies target selection rather than mask decoding as the main bottleneck. On a controlled training subset, matched supervision improves identity-aware performance, showing that the diagnosed capability responds to targeted supervision. Collectively, InstanceBench supports a measure-diagnose-improve cycle: measuring target consistency across grounding paths, localizing failure sources, and evaluating targeted interventions.

cs.CV↗

arXiv2026

PuzzleJAX: A Benchmark for Reasoning and Learning

We introduce PuzzleJAX, a GPU-accelerated puzzle game engine and description language designed to support rapid benchmarking of tree search, reinforcement learning, and LLM reasoning abilities. Unlike existing GPU-accelerated learning environments that provide hard-coded implementations of fixed sets of games, PuzzleJAX allows dynamic compilation of any game expressible in its domain-specific language (DSL). This DSL follows PuzzleScript, which is a popular and accessible online game engine for designing puzzle games. In this paper, we validate in PuzzleJAX several hundred of the thousands of games designed in PuzzleScript by both professional designers and casual creators since its release in 2013, thereby demonstrating PuzzleJAX's coverage of an expansive, expressive, and human-relevant space of tasks. By analyzing the performance of search, learning, and language models on these games, we show that PuzzleJAX can naturally express tasks that are both simple and intuitive to understand, yet often deeply challenging to master, requiring a combination of control, planning, and high-level insight.

cs.AI↗

arXiv2026

Your Unlearning Gives You Away: Identifying Erased Concepts in Diffusion Models

Existing attacks on unlearned diffusion models assume that the erased concepts are known in advance and focus on recovering them. In practice, however, model providers may not disclose which concepts have been removed, and even with access to the original base model, an adversary may still lack a clear target to attack. In this paper, we aim to answer the following critical but overlooked questions: which concepts have been erased from the model, and how many have been erased in total? To this end, we present Tracer, a framework that rapidly and accurately identifies erased concepts and estimates their number. Tracer efficiently identifies erased concepts without generating and classifying images. By combining lightweight spectral analysis of weight footprints, it enables efficient search over large candidate vocabularies. To distinguish multiple erased concepts, we introduce a footprint coverage objective that guides sequential discovery. Tracer estimates the number of erased concepts by detecting a sharp decline in candidate confidence as the selected concepts account for the erasure footprint, without requiring labeled examples for calibration. The framework requires only lightweight linear algebra and limited forward probes, with no prior knowledge of the unlearning algorithm. Experiments across text-to-image and text-to-video backbones and diverse unlearning methods demonstrate that Tracer identifies erased concepts and estimates their number in seconds, achieving 150 to 137,000 times and 133 to 20,000 times speedups over MIA and brute-force search on image and video models, respectively, with substantially higher identification accuracy.

cs.LG↗

arXiv2026

OpenSpace Lab Solution to the IROS 2026 Indoor Exploration Competition

This report presents the \textbf{OpenSpace Lab}'s solution to the Competition on Intelligent Information Gathering for Single and Multi-Robot Systems Workshops, organized as part of the IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) 2026. Our team reached 1st place in the Single-Robot Public Track and 3rd place in both the Single- and Multi-Robot Private Tracks. The single-robot framework utilizes pre-trained map completion predictions for global planning to prioritize unexplored areas. To reconcile map coverage with limited operation time, we introduce a remaining-time-based exploration strategy that integrates homing constraints into the decision-making process. For multi-robot exploration, we utilize a utility-driven target selection strategy that balances observation gains, movement costs, and budget constraints, leveraging shared map and intent data to eliminate redundant search and maximize coordination efficiency. Our solution reached a 61.04\% coverage rate in the Single-Robot Public Track, while reaching 39.53\% and 39.91\% coverage in the Single- and Multi-Robot Private Tracks, respectively. An extended full-length paper based on this report is currently being prepared for submission, and the source code will be released upon acceptance of the full manuscript at https://github.com/OpenSpace-Lab/Indoor-Exploration-IROS2026.

cs.RO↗

arXiv2026

Routing in Gradient Space: Balanced Usage Is Not Expert Specialization

Sparse expert models can distribute traffic evenly while still grouping incompatible training signals within the same experts. We study routing as a gradient-partitioning problem and introduce gradient-aligned routing (GAR), whose load-normalized router objective rewards grouping observations with aligned gradients. On five multi-task text-classification mixtures, we compare GAR with task-loss-only routing, gradient-combination and gradient-conflict methods, and load-balancing losses. With a fully trainable RoBERTa backbone and classification-head experts, GAR has the highest aggregate validation accuracy, 1.07 percentage points above task-loss-only routing. With frozen DeBERTa and Qwen3-1.7B backbones and low-rank adapter experts, it again ranks first, 1.10 points above task-loss-only routing, with better-balanced expert load and higher gradient-mass purity, the share of each expert's gradient-norm mass from its dominant task; the load-balancing losses flatten load further but leave this purity near its task-loss-only level. Top-1 routing, trainable full-parameter feed-forward network (FFN) experts, and a larger backbone also show positive aggregate gains. The results distinguish expert-load balance from gradient-based routing organization and indicate the predictive value of gradient-informed routing in multi-task text classification.

cs.LG↗

arXiv2026

Backpropagated Output Momentum: Relocating Optimizer History from Parameters to Task Space

Optimizer momentum is usually stored as a parameter-sized moving average of past gradients, which makes history costly and fixes each past signal in the coordinates in which it was computed. We introduce Backpropagated Output Momentum (BOM), which instead stores a compact moving average of prediction errors at the model output and reprojects that history through the current network at every step. A batch-level analysis characterizes the information retained and omitted by this relocation, while the implementation preserves the current supervised gradient and can replace the first-moment component of several adaptive optimizers. As a plug-in for momentum-based optimizers, including ones that already compress their state, BOM reduces parameter-shaped optimizer state by 49.7-99.8% in three compositions and, averaged over three language backbones, paired step time by 4.0%. It also improves mean validation performance across language and vision fine-tuning, by 1.42 points in the primary five-task comparison. Language and vision pretraining studies, together with matched mechanism controls, further test the construction across output spaces and model scales.

cs.LG↗

arXiv2026

QAM: Quadratic-Accurate Checkpoint Merging via Sequential Consistency

Saved checkpoints record states along a training trajectory, but generally do not determine the updates at states that would be visited under a different schedule. We study how accurately these checkpoints can reconstruct the endpoint of a sequential reference with prescribed update strengths. Under a common local transition model, two checkpoint-index moment conditions characterize all convex merges that agree with this reference through second order. We then prove an information limit that for nondegenerate profiles, no algorithm using only a fixed-length gradient-descent (GD) history with step size $h$ can achieve $o(h^3)$ endpoint error uniformly over a fixed class of smooth, strongly convex losses. The lower bound follows from two losses with identical GD checkpoint histories but sequential reference endpoints separated by $Ω(h^3)$. \textbf{Quadratic-Accurate Merging} (QAM) achieves a matching uniform $O(h^3)$ endpoint error bound. Its explicit coefficients also define the unique profile-dependent merge that exactly matches the sequential GD reference across all fixed quadratic objectives. Across two public Adam checkpoint trajectories (SmolLM3-3B and OpenEuroLLM-Prelude-9B), three windows and three profiles per model, and 15 tasks, QAM shows mixed results for short windows and broader advantages over \textbf{Warmup-Stable and Merge} (WSM) for longer windows. Matched-moment GSM8K diagnostics further show that local consistency alone does not fully determine downstream scores. These results characterize the reconstruction limits of saved histories, provide a coefficient rule that attains the optimal rate, and assess its practical utility.

cs.LG↗

arXiv2026

Benchmarking Tabular Foundation Models as Surrogates in Expensive Evolutionary Optimization

Surrogate-assisted evolutionary algorithms (SAEAs) are effective methods for solving expensive optimization problems (EOPs), where surrogate models replace most expensive evaluations and critically influence the final optimization results. In recent years, tabular foundation models have advanced rapidly, and the Tabular Prior-data Fitted Network (TabPFN) has been adopted as a surrogate model for EOPs due to its strong predictive capability, demonstrating promising performance. Motivated by its potential as a surrogate model in SAEAs, this work conducts a comprehensive study that combines extensive experiments with in-depth theoretical analysis to investigate the effectiveness of TabPFN. Specifically, we perform experiments across both offline and online SAEA settings, covering diverse problem scenarios such as single-objective, multi-objective, constrained, combinatorial, mixed-variable, and engineering optimization problems. In addition, we further analyze the advantages and limitations of TabPFN within SAEAs and provide practical guidelines for its application in different optimization settings. Results show that the effectiveness of TabPFN is highly problem dependent, and it cannot replace conventional surrogates universally. Overall, TabPFN should be adopted selectively according to data availability, landscape complexity, search space characteristics, and its role within the algorithm. Customized model management strategies and role-specific algorithm design are necessary to fully exploit its advantages and avoid its pitfalls.

cs.NE↗

arXiv2026

X-Stage: Modeling Post-Issue Backpressure in GPU Communication--Computation Fusion

Fine-grained, device-initiated communication allows fused GPU kernels to issue remote stores directly from their compute pipelines, a pattern increasingly used in expert parallelism (EP), tensor parallelism (TP), and Ulysses-style sequence parallelism (UP). Existing designs reason about where communication is issued and when remote data becomes ready, but lack a quantitative model of the sender-side interval after a remote store is accepted and before it becomes visible at the destination. This interval determines whether communication remains decoupled from computation or backpressures it. We identify X-Stage, a software-visible post-issue stage with finite decoupling. Downstream pressure can dissipate while the issuer resumes useful work, whereas sustained injection consumes X-Stage headroom and eventually stalls the compute pipeline. We characterize this behavior and build a calibrated model that predicts whether remote-store arrivals accumulate backpressure or recover during intervening computation. Guided by the model, we reshape bursty arrivals when they would exhaust X-Stage headroom and exploit natural compute windows when headroom can recover concurrently. Evaluation across representative EP, TP, and UP workloads shows up to 1.62x fused-kernel, 1.75x end-to-end, and 1.43x sender-visible speedup, respectively. Microbenchmarks further validate the model's predictions of backlog accumulation, recovery, and sender-side backpressure.

cs.DC↗

arXiv2026

SWRouter: Similarity-Contractive Window Routing for Multi-Turn Large Language Model Conversations

Large language models exhibit complementary strengths, motivating routing methods that dispatch each query to the most suitable model. Although existing routers are effective in single-turn settings, they do not directly transfer to multi-turn dialogue, where routing performance critically depends on how historical context is segmented, retained, and incorporated into the current prompt. This introduces two fundamental challenges: preventing information loss and information confusion during context construction, and evaluating routing quality without conflating model selection with prompt construction quality. In this paper, we propose SWRouter, a Similarity-Contractive Window Router for multi-turn large language model routing. SWRouter combines a similarity-based context segmentation mechanism for prompt construction with a dual-metric evaluation framework that decouples construction accuracy from router performance. Experiments on multi-turn dialogue benchmarks demonstrate that SWRouter consistently surpasses strong baselines, achieving a 16.26% improvement in evaluation accuracy over the best individual large language model and an additional 8.22% gain over the Conv-ID Context baseline. Our results highlight that multi-turn large language model routing requires a joint design of context construction and evaluation, rather than a direct extension of single-turn routing methods.

cs.CL↗

arXiv2026

What Should an Agent Forget? Separating What Is Stored from What Is Used

Persistent language agents need stored experience to remain available across time, while each answer requires evidence suited to a particular question. A superseded fact can mislead a current-state answer and still be essential for a historical query. We present RD-Forget, a training-free framework that separates what an agent stores from what it uses. A retained source archive preserves observations, and a query-conditioned memory view controls their influence on the current answer. A frozen language-model curator extracts relevant evidence, groups facts into semantic slots, and preserves the relations needed for multi-hop reasoning. Same-slot replacement links suppress superseded values in current-state contexts, while intent-aware retrieval makes earlier evidence eligible again. A rate-distortion formulation guides construction of the answer-time view within a memory budget. Experiments span conversational memory, knowledge updating, fact consolidation, long-context reasoning, and personalization under a shared answering pipeline. The results associate accurate answers with both query-relevant evidence construction and control over obsolete alternatives. Configurations without forgetting or query conditioning have the largest score deficits, while slot grouping, historical access, and relation preservation contribute complementary functions. Retaining history while selectively controlling its use offers a practical way to accommodate changing facts and future questions.

cs.AI↗