Searcharxiv⌕ Search

arXiv subjects

Yi Li

Publications and source records attributed to Yi Li.

At least 55 records · Page 3Linked to original sources
arXiv2026

Targeted Deep Survival Contrasts: Valid Inference for Treatment-Specific Survival Benefit with Neural Networks

Neural survival models are increasingly asked to support counterfactual claims---how much a treatment would change survival in a population---rather than only prognostic risk scores. Answering such questions from observational data requires valid inference for treatment-specific survival contrasts under confounding and covariate-dependent censoring, targets for which standard deep survival estimators are biased and provide no honest uncertainty. We propose Targeted Deep Survival Contrasts (TDSC), which extends Targeted Deep Architectures (TDA)---targeted maximum likelihood estimation embedded in a network's weight space---to the full vector of treatment-specific survival curves over a time grid, and hence to the benefit curve and the restricted mean survival time (RMST) difference. A single universal targeting path, one ridge projection of the stacked efficient influence functions onto closed-form last-layer gradients per iteration, simultaneously solves the projected estimating equations for all coordinates; a one-step residual top-up converts the plug-in into a doubly robust estimator of the unrestricted target; and a multiplier bootstrap yields simultaneous confidence bands for the benefit curve. We prove joint asymptotic linearity, band validity, and double robustness of the top-up for a cross-fitted variant requiring no Donsker conditions. Across seven Monte Carlo banks with confounded treatment, sign-varying effect heterogeneity, and dependent censoring, the TDSC plug-in attains nominal pointwise and simultaneous coverage with 35% lower MSE than a per-timepoint one-step (AIPCW) built from the same nuisance fits. Under a badly wrong outcome model the plug-in tracks its working parameter and its intervals fail (44% coverage), while the top-up restores nominal inference for the unrestricted causal target (94-95%)---and in-sample diagnostics separate the two regimes.

stat.ME↗
arXiv2026

CACSurv: Concordance-Aligned Comparative Learning with Large Language Models for Cancer Survival Prediction

Cancer survival prediction supports treatment planning, risk stratification, and follow-up management. Existing methods use structured clinical variables, whole-slide images, genomic profiles, or multimodal inputs, while patient reports remain underexplored. We study report-centric survival prediction using reports that organize pathological, clinical, and molecular evidence. Large language models (LLMs) can reason over such reports, but case-wise time regression introduces two mismatches. First, a formulation mismatch arises because survival evaluation depends on ordering comparable patients, whereas independent time predictions do not enforce ranking consistency. Second, a supervision mismatch arises because a censored patient's observed time indicates survival beyond that point and cannot serve as an exact regression target, although it still implies orderings relative to patients who died earlier. To address these mismatches, we propose CACSurv, a Concordance-Aligned Comparative framework for report-centric survival prediction. CACSurv reformulates survival modeling as mini-cohort comparative reasoning, where an LLM predicts relative prognostic orderings. We introduce concordance-aligned rewards derived from comparable relations under right censoring, enabling censored outcomes to provide ranking supervision without exact event-time targets. At inference, Monte Carlo Reference Aggregation compares each patient with sampled references and aggregates positions into a cohort-level ranking. We establish TCGA-SurvReport, a benchmark covering six TCGA cancer cohorts. CACSurv achieves the highest C-index on all six cohorts and an average C-index of 0.722, outperforming the strongest published survival model by 6.5 percentage points and the strongest LLM time-regression baseline by 4.2 percentage points. Our code, models, and dataset will be available at https://github.com/xmed-lab/CACSurv.

cs.AI↗
arXiv2026

Degraded Infrared Small Object Detection via Degradation-Adapted Physics-Guided Restoration

Infrared small object detection has made significant progress in recent years. However, degradations such as fog and nonuniformity can suppress target-background contrast, substantially increasing detection difficulty. Existing methods mainly rely on image restoration as preprocessing, but they are typically designed for specific degradation types and fail to generalize to varying degradations. To alleviate this, we propose DAISOD, a degradation-adapted infrared small object detection framework for robust detection under different degradations. DAISOD first identifies the type and severity of degradations, then adapts the processing via dedicated branches, and finally fuses the results for subsequent detection. Moreover, a physics-guided restoration mechanism is incorporated to explicitly estimate degradation parameters and remove degradation effects through physical models, avoiding excessive restoration that may erase small targets. Moreover, we construct a degraded infrared small object detection dataset covering diverse degradation types and levels. Extensive experiments show that DAISOD outperforms state-of-the-art methods under various degradation conditions.

cs.CV↗
arXiv2026

A Near-Optimal Lower Bound for $\ell_p$-Subspace Embeddings, $1\leq p<2$

For $d \geq 2$, $p \geq 1$ and $ε> 0$, let $N_p(d,ε)$ be the smallest integer $N$ such that for every integer $n$ and every $A\in\mathbb{R}^{n\times d}$, there exists a matrix $Φ\in\mathbb{R}^{N\times n}$ satisfying $(1-ε)\lVert Ax\rVert_p\leq \lVertΦA x\rVert_p\leq (1+ε)\lVert Ax\rVert_p$ for all $x\in\mathbb{R}^d$. For every constant $p\geq 1$ with $p\not\in 2\mathbb{Z}$, when $d\gtrsim_p \log(1/ε)$, the bound \[ N_p(d,ε) \gtrsim_{p} \frac{d}{ε^2 \operatorname{polylog}(d/ε)} \] is established. This improves the previous lower bound $Ω(1/(ε^2\operatorname{polylog}(1/ε)))$ due to Li et al. (SICOMP 2021) and is optimal up to logarithmic factors for $1\leq p<2$. The central technical idea originated from ChatGPT 5.6 Sol.

cs.DS↗
arXiv2026

KHiM-Mamba: Injecting Pathology Knowledge into Mamba via Hidden-State Modulation for Whole Slide Image Analysis

Whole slide image analysis is commonly formulated as multiple instance learning (MIL), where instance features are contextually updated and aggregated into a slide representation, a process we term slide encoding dynamics. Recently, selective state-space models (SSM) have emerged as promising MIL architectures due to their long-sequence modeling capability and linear complexity. However, existing SSM-based MIL methods rely solely on visual features during MIL. Meanwhile, in large-scale WSIs, where sparse diagnostically decisive regions are surrounded by abundant irrelevant information, such purely vision-driven selective dynamics can misallocate state updates and readouts, causing the evolving SSM state to accumulate task-irrelevant evidence and dilute critical diagnostic cues over long scan trajectories. In this work, we propose the Knowledge-Aware Hidden-State Modulation architecture (KHiM-Mamba), which innovatively regulates Mamba's core selective state-space mechanism with explicit knowledge priors, steering slide encoding dynamics toward diagnostically meaningful evidence accumulation. Specifically, we redesign the original SSM layer to perform knowledge modulation operations during the evolution of hidden states, thereby guiding what visual evidence is accumulated and retrieved from the hidden state at each encoding step. Furthermore, we additionally introduce a local-adaptive vocabulary retrieval module that uses large language models to assign each patch fine-grained, tissue-specific semantic descriptions, enabling precise modulation across diverse tasks. Experiments on 11 public benchmarks across 4 tasks show that KHiM-Mamba consistently achieves state-of-the-art performance.

eess.IV↗
arXiv2026

Decoding Mixture Perception through Computational Modeling of Component Interactions

Olfaction played an indispensable role throughout human evolution and civilization. Even in the contemporary era of advanced technology, olfaction remains a critical channel for person to conduct danger discrimination, emotional experience, and memory formation. However, most substances in nature exist as multi-molecule mixtures. The complexity of mixture compositions, as well as concentration dependent saturation effects and receptor specific activation thresholds, pose substantial challenges in identifying olfactory characteristics. In this study, we proposed a novel bio inspired deep learning framework for accurate odor perception recognition of mixtures. We robustly constructed neural response curves for molecule-receptor interactions, and developed a fusion strategy that integrates attention-weighted multi-receptor curves with concentration-dependent multi-molecule curves, replicating the competitive activation and synergistic integration of mixture components. Furthermore, by comparing the consistency of response curve patterns, the model can transfer knowledge from the semantically rich space of molecular associations to guide recognition of mixture perception characteristics. Therefore, we established a complete computational pathway from chemical blending, neural encoding, to perceptual formation. Finally, we conducted comprehensive evaluation, and results demonstrated exceptional superiority, achieving an accuracy of 92.2%. Consequently, our work provides a generalizable solution to the long standing mixture perception challenge. More importantly, it can be integrated into embodied cognitive systems to enhance the agents perceptual and interactive capabilities in complex scenarios.

cs.LG↗
arXiv2026

Mechanism-Dependent Descriptors Enable Predictive Design of Oxygen Capacity in Perovskite Oxides

Perovskite oxides can reversibly accommodate substantial changes in oxygen stoichiometry, making them attractive for clean-energy technologies including chemical looping and oxygen storage. Despite extensive efforts to optimize their redox properties, predictive descriptors capable of assessing oxygen capacity across diverse compositions remain under development. Here, we combine experiments and first-principles calculations to establish composition and oxygen-capacity relationships in the model perovskite series LnxSr1-xCoO3. We confirm that increasing Sr2+ content promotes the formation of high-valence Co4+, expanding the cationic redox reservoir available during oxygen release and thereby enhancing oxygen capacity. In this regime, oxygen-vacancy formation energy captures the observed trend because oxygen release is primarily compensated by Co4+/Co3+/Co2+ redox. Across the rare-earth series, however, oxygen capacity decreases from La to Lu despite progressively lower oxygen-vacancy formation energies. We reveal that this counterintuitive behavior originates from an alternative charge-compensation pathway, in which lattice oxygen is partially oxidized to O1- -like species during oxygen removal. Heavy rare-earth compositions (Tb-Lu) preferentially stabilize these oxygen-hole species through distinct local bonding environments, with charge compensation involving both oxidized lattice oxygen and reduced rare-earth and cobalt cations, thereby suppressing net oxygen release despite favorable vacancy thermodynamics. We further identify average metal-oxygen bond strength, quantified by integrated crystal orbital Hamilton population, as a physically meaningful descriptor for oxygen capacity when anionic redox becomes dominant.

cond-mat.mtrl-sci↗
arXiv2026

Autonomous Optimization of Complex Oxides for Thermochemical Fuel Production

Two-step thermochemical fuel production, including H2O and CO2 splitting, offers a promising route to sustainable fuel manufacturing, with performance governed by redox-active oxides that enable cyclic reduction-oxidation reactions. Maximizing thermal-to-fuel conversion efficiency demands materials that simultaneously satisfy multiple stringent thermodynamic and kinetic targets. Addressing these requirements has increasingly driven materials design toward complex, multi-cation oxides, such as mixed-cation fluorites, perovskites, and high-entropy oxides, wherein composition, defect chemistry, phase stability, and morphology should be co-optimized. This creates a challenging materials optimization problem that is poorly suited to traditional trial-and-error approaches. In this review, we argue that thermochemical fuel production provides a compelling frontier for autonomous materials design and optimization. We first examine why redox-active complex oxides are difficult to develop, owing to multidimensional phase spaces, harsh operating conditions, and competing functional targets. We then discuss how high-throughput computation, automated synthesis, characterization and testing, and machine learning can be integrated into closed-loop workflows to address these challenges. Building on broader oxide materials research, we organize recent progress into a capability roadmap for complex-oxide optimization, spanning compositionally diverse synthesis, operando characterization, robotic testing, operation-condition computation, and multi-objective optimization. Finally, we outline key experimental, computational, and data challenges for building self-improving materials development platforms for materials development in thermochemical fuel production.

cond-mat.mtrl-sci↗
arXiv2026

Knowledge-Driven Hybrid SSD Management Enhanced by Fine-Tuned LLMs

A hybrid Solid-State Drives (SSDs) integrates different modes of flash cells (e.g., single-level cell (SLC) and Quad-Level Cell (QLC)) and enables them to convert between each other, achieving both high performance and storage capacity. However, this hybrid design introduces a significantly larger design space than traditional SSDs with additional design factors such as flash conversion and data migration across different flash modes, leading to higher optimization complexity. Efficient management of such complexity requires deep hybrid SSD knowledge and dynamic adjustment mechanisms. Large language models (LLMs) offer a promising solution through their contextual reasoning and adaptive coordination capabilities. In this work, we explore the potential of using LLMs in understanding and efficiently managing hybrid SSD design space. We find that leveraging LLMs for knowledge-guided optimization of management parameters enables substantial performance gains. Building on these insights, we propose LLM-hybridSSD, an integrated optimization framework that formulates hybrid SSD management as a parameter-tuning problem, employs an LLM-based tuner for adaptive configuration, and applies reinforcement learning-based fine-tuning to align local lightweight models with domain-specific knowledge. Experimental results show an average 58.92% increase in throughput and a 28.56\% reduction in write amplification (WA) compared with state-of-the-art schemes under different real-world workloads.

cs.AR↗
arXiv2026

Combating Knowledge Corruption in Agent Systems: A Byzantine-Tolerant Secure Collaborative RAG Framework

While retrieval-augmented generation systems partially address the hallucination issues in large language models, it also introduces new vulnerabilities to knowledge corruption attacks. Adversaries exploit these vulnerabilities by poisoning documents provided by RAG system to manipulate LLM outputs. To counter this threat, we propose SecureCollaRAG, a Byzantine-tolerant collaborative RAG framework leveraging Multi-source Knowledge Validation Mechanism. Our approach enables agent system to securely verify document provenance through dynamic GNN-based credibility scoring, effectively preventing stealthy knowledge corruption attacks while preserving essential domain knowledge integrity. Through extensive evaluations and formal analysis, we demonstrate that SecureCollaRAG maintains robustness against attackers under non-IID data distributions.

cs.CR↗
arXiv2026

Recent Sharp Rise in Inhomogeneous Hydrological Extremes Stress Vegetation Growth in China

The intensifying spatial inhomogeneity of rainfall under greenhouse warming implies that more inhomogeneous hydrological extremes (IHEs), i.e., coexistence of extreme rainfall or drought, may be triggered. While vegetation growth in China is sensitive to hydrological hazards, the variability of IHEs and their ecological impacts remain underexplored. Here, we find a significant increase in IHEs during China's growing season since 2000 (+2.1 events or +14.52 days per decade), with a rapid sharp rise to an annual average of 6.4 events or 42.0 days in the past five years. The primary driver is the enhanced inhomogeneity of moisture-dynamic coupled weather conditions, overlapping with a northward shift of climatological precipitation distribution. This results in a "Wet-North and Dry-South" pattern of IHE impacts, which poses severe and asymmetric threats to vegetation growth in China, with the expansion of drought areas exerts stronger stress on vegetation than the compensatory effects of rainfall. Our findings suggest that the sharp rises of IHEs tend to yield net negative impacts on vegetation growth, highlighting the need for stronger hydrological management to reduce future risks.

physics.ao-ph↗
arXiv2026

SUV: Future Scene Understanding as Video Generation for End-to-End Driving

End-to-end driving requires a coherent understanding of future scenes, yet existing methods model these scenes using task-specific heads and output formats, with limited scalability. Can video generation instead provide a shared predictor? We introduce SUV, a unified end-to-end driving framework that casts future Scene Understanding as Video generation using a pretrained video foundation model. SUV models future appearance, semantics, relative depth, and instance-level dynamics as video streams with a shared video expert, without stream-specific visual prediction heads. Through joint video-action attention, the action expert attends to the latent representations of all future streams and generates the ego trajectory. Experiments show that SUV directly predicts all four future streams, while controlled ablations show that structured future supervision and direct future-stream access yield higher trajectory planning scores. With only a single front camera and no candidate-trajectory selection, SUV outperforms a broad set of recent state-of-the-art methods on both NAVSIM-v2 splits, achieving 91.0 EPDMS on navtest and 36.9 on navhard. On the long-tail WOD-E2E benchmark, SUV achieves a competitive RFS of 7.94.

cs.CV↗
arXiv2026

MoWorld: A Flash World Model

The future of World Models depends not only on scaling model capability, but also on scaling practicality and inference efficiency. High-frame-rate inference enables responsive perception, planning, and control in real-world autonomous systems. To this end, we present MoWorld, a cost-effective yet high-performance Flash World Model with an end-to-end framework spanning data generation, pre-training, distillation, and efficient inference, enabling up to 50 FPS real-time interaction with cinematic visual quality without the need of high-end GPUs. To enable large-scale real-world deployment, MoWorld jointly optimizes model capability and cost throughout the entire development pipeline. Specifically, unlike existing approaches that primarily rely on large-scale video corpora, MoWorld is built upon a scalable 3D-native data engine accumulated from our large-scale 3D vision and generative modeling pipeline, enabling the efficient construction of geometrically consistent training data across diverse real-world and synthetic environments. Based on this foundation, a curriculum cross-frame pre-training strategy for stable and scalable World Model learning, an efficient denoising-step distillation algorithm to reduce diffusion training cost, and a mixed-precision parallel inference framework for low-cost real-time deployment. MoWorld is the first real-time interactive World Model built on the Neural Processing Unit (NPU) and can achieves up to 50 FPS in such the devices, enabling practical and efficient deployment at scale. Comprehensive evaluations demonstrate that MoWorld achieves leading performance; notably, its average inference cost is only 30\%-50\% of that of existing World Models, providing a practical foundation for large-scale real-world applications of World Models. We also demonstrate diverse applications of MoWorld.

cs.CV↗
arXiv2026

Optimal Embeddings of Constant-Dimensional Subspaces of $L^p$ into $\ell_p^N$

For $d \geq 2$, $p \geq 1$ and $ε> 0$, let $N_p(d,ε)$ be the smallest integer $N$ such that every $d$-dimensional subspace of $L^p[0,1]$ admits a linear embedding into $\ell_p^N$ with distortion at most $1 + ε$. For fixed $d\geq 2$ and $p\geq 1$, the bound \[ N_p(d,ε) \lesssim_{d,p} ε^{-2(d-1)/(d+2p)} \] is established. For $p \notin 2\mathbb{Z}$, this matches the known lower bound up to constant factors. For odd integers $p$, previous upper bounds with this exponent incurred additional logarithmic factors, except in the logarithm-free case $p = 1$; for non-integral $p$, no upper bound with this exponent was previously known. For even integers $p$, isometric embeddings of dimension independent of $ε$ are known. For $p \notin 2\mathbb{Z}$, the proof approximates $|t|^p$ by a polynomial with a remainder of small total variation. The polynomial part contributes no error, while the error from the remainder is controlled by an integrated equatorial-band discrepancy estimate.

math.FA↗
arXiv2026

Latency-Optimal Adaptive Split Inference for Privacy-Preserving Cloud-Edge-End Collaboration

Internet of Things (IoT) end devices are increasingly expected to support privacy-sensitive batch inference, yet their limited computational resources often make full local execution of convolutional neural networks impractical. This paper presents a latency-optimal adaptive split inference framework for privacy-preserving cloud-edge-end collaboration. The end device acts as the trust anchor, executes the plaintext model prefix, encrypts the split activation using fully homomorphic encryption (FHE), and keeps the secret key locally, while the edge and cloud execute assigned model segments only on FHE ciphertexts. We formulate collaborative encrypted inference as a split-pair selection problem over an end-side split point and an edge-side termination point. The proposed planner jointly models plaintext prefix execution, encryption, communication, edge-side FHE execution, and cloud-side FHE completion, and supports both convolution-level and block-level split granularities. Experiments on CIFAR-10 and PathMNIST show that the proposed convolution-level collaborative scheme achieves amortized end-to-end speedups of approximately 12.9 times over full-cloud FHE and 3.9 times over the block-level alternative, while preserving the corresponding plaintext-model accuracy. Including modeled communication, the amortized latencies are 1033.279 s/sample on CIFAR-10 and 1023.429 s/sample on PathMNIST.

cs.CR↗
arXiv2026

Active Regression for Single-Index Models with Unknown Link Functions

This paper studies active regression for single-index models under general $\ell_p$-loss with an unknown $1$-Lipschitz link function $f$, formulated as $\min_{f,x} \|f(Ax)-b\|_p^p$ with full access to $A$ but coordinate-query access to $b$. Prior work established upper bounds for known link functions for all $p\geq 1$ and for unknown link functions only in the $p=2$ case, together with lower bounds for $p\leq 2$. This work addresses the more challenging setting of unknown link functions and general $p \geq 1$. A non-adaptive sampling algorithm is presented that achieves a $(1+ε)$-approximation using $O(d^{p/2\vee 1}/ε^{p\vee 2}\operatorname{poly}\log(n/ε))$ queries. Nearly tight lower bounds are also established for $p>2$. These results close much of the remaining gap in active $\ell_p$-regression for single-index models.

cs.DS↗
arXiv2026

V-Mem: Modality-Routed Retrieval for Long-Term Multimodal Agentic Memory

Interaction between users and LLM agents is increasingly multimodal: conversations interleave text with images, and a later question may target either. Yet most agent memories are designed around text, and even the few that support multimodal conversations still fail on vision-related questions. We trace this failure to an assumption behind the similarity search they rely on: in the index space, a query lies close to the relevant evidence that answers it. In multimodal settings, two gaps break it. By the modality gap, a query lies closer to memory content of its own modality than to evidence in another, even in a trained joint embedding space. By the similarity-relevance gap, the content most similar to a query is often not the evidence that answers it, most acutely when a query carries both text and image and its evidence resembles neither part alone. We present V-Mem, a multimodal agentic memory system that routes retrieval by the modality of the query and that of the target evidence, both recognized from the query alone. To cross the modality gap, V-Mem organizes the conversation into rounds and returns the target-modality content from the same round as the match, without comparing across modalities. To close the similarity-relevance gap, it searches with an LLM-generated anchor that sits closer to the relevant evidence than the query does: a hypothetical caption for a text-only query seeking an image, and an enriched search anchor, the query text plus relevant keywords extracted from the query image, when the evidence is reachable only by combining the two. On Mem-Gallery, V-Mem reaches an LLM-judge score of 0.82 versus 0.56 for the second best, with the largest margin on questions carrying an image (0.87, no baseline above 0.47); on LoCoMo it scores 0.69 versus 0.58.

cs.AI↗
arXiv2026

UniVA: Unified Value Alignment for Generative Recommendation in Online Advertising at Tencent

Generative Recommendation (GR) reformulates recommendation as next-token generation over item Semantic IDs (SIDs) and has shown promise in industrial applications. However, extending GR to advertising is non-trivial, as advertising recommendation must jointly account for user relevance and commercial value. This creates a mismatch: high generation likelihood does not necessarily imply high advertising utility. As a result, valuable ads may be poorly distinguished in the SID space, pruned during autoregressive decoding, or missed when request-invalid branches consume limited beam capacity during online serving. To address this problem, we propose UniVA, a Unified Value Alignment framework for generative advertising recommendation. UniVA aligns commercial value across the entire pipeline of SID construction, autoregressive decoding, and online serving. Commercial SID Tokenization injects business attributes and bid information into SID construction. A Generation-as-Ranking SID Decoder then fuses generation scores with token-level value estimates during autoregressive decoding. {Finally, Value-Aware Constrained Serving restricts the fused decoding process to request-valid SID paths through a personalized trie.} Experiments on the Tencent WeChat Channels advertising platform show that UniVA achieves a 37.04\% relative improvement in offline Hit Rate@100 over the baseline and lifts gross merchandise value (GMV) by 1.5\% in online A/B tests.

cs.IR↗