arXiv subjects

Tianqi Chen

Publications and source records attributed to Tianqi Chen.

At least 19 recordsLinked to original sources

arXiv2026

AnyCXR: Human Anatomy Segmentation of Chest X-ray at Any Acquisition Position using Multi-stage Domain Randomized Synthetic Data with Imperfect Annotations and Conditional Joint Annotation Regularization Learning

Robust anatomical segmentation of chest X-rays (CXRs) remains challenging due to the scarcity of comprehensive annotations and the substantial variability of real-world acquisition conditions. We propose AnyCXR, a unified framework that enables generalizable multi-organ segmentation across arbitrary CXR projection angles using only synthetic supervision. The method combines a Multi-stage Domain Randomization (MSDR) engine, which generates over 100,000 anatomically faithful and highly diverse synthetic radiographs from 3D CT volumes, with a Conditional Joint Annotation Regularization (CAR) learning strategy that leverages partial and imperfect labels by enforcing anatomical consistency in a latent space. Trained entirely on synthetic data, AnyCXR achieves strong zero-shot generalization on multiple real-world datasets, providing accurate delineation of 54 anatomical structures in PA, lateral, and oblique views. The resulting segmentation maps support downstream clinical tasks, including automated cardiothoracic ratio estimation, spine curvature assessment, and disease classification, where the incorporation of anatomical priors improves diagnostic performance. These results demonstrate that AnyCXR establishes a scalable and reliable foundation for anatomy-aware CXR analysis and offers a practical pathway toward reducing annotation burdens while improving robustness across diverse imaging conditions.

cs.CV↗

arXiv2026

Let the Gate Decide: Per-Lookup Precision for Quantized Conditional Memory in LLMs

Conditional memory layers, such as DeepSeek's Engram, attach large hashed n-gram tables to a language model, and removing them from DeepSeek-V4.1-Flash costs 36.8 PopQA points. Low-bit quantization loses long-tail facts first, and the standard remedy, giving more bits to frequently accessed rows, backfires: refining lookups by access frequency is worse than refining them at random. We trace this to the gate that scales each retrieved row. A lookup's quantization error is weighted by its gate, which is small for frequent n-grams, large for the rare n-grams of long-tail entities, and computable from a 1-bit code. GateRQ therefore stores each row as a ladder of 1-bit residual codes, keeps level 1 in GPU memory to compute the gate, and reads deeper levels only for the lookups the gate ranks highest. Without retraining, GateRQ cuts GPU table memory by 7.8x on DeepSeek-V4.1-Flash, which raises vLLM throughput by 11%-31% over FP8 tables, while losing about 1 PopQA point at 1.50 bits read per lookup and 0.2 points at 2.10 bits across two models.

cs.LG↗

arXiv2026

RAQ-MLA: Role-Aware Quantization for Multi-Head Latent Attention Caches

Multi-Head Latent Attention (MLA) has been adopted by recent frontier large language models. Instead of per-head keys and values, MLA caches a compressed KV latent that is up-projected into keys and values at decode time, together with a narrow positional key that carries position information into the attention score. Existing KV-cache quantization methods are designed around the keys and values of standard attention and do not account for what MLA actually caches, which costs accuracy at low bit widths. We analyze the MLA cache both in activation distribution and in how quantization error propagates to the attention output, and show that the two cached tensors play different roles: the latent mainly supplies the values, whereas the positional key dominates the attention score. We therefore propose RAQ-MLA (Role-Aware Quantization for Multi-Head Latent Attention Caches), which fits an invertible transform to each cached tensor against the error that tensor delivers to the attention output: attention-weighted value error for the latent and score perturbation for the positional key. Without training, RAQ-MLA quantizes the full MLA cache to about 2.5 bits per element, a 6.5x increase in theoretical KV-cache capacity, while remaining nearly lossless on reasoning and coding tasks across four MLA models.

cs.AI↗

arXiv2026

TernaryLLM: Ternarized Large Language Model

Large language models (LLMs) have achieved remarkable performance on Natural Language Processing (NLP) tasks, but they are hindered by high computational costs and memory requirements. Ternarization, an extreme form of quantization, offers a solution by reducing memory usage and enabling energy-efficient floating-point additions. However, applying ternarization to LLMs faces challenges stemming from outliers in both weights and activations. In this work, observing asymmetric outliers and non-zero means in weights, we introduce Dual Learnable Ternarization (DLT), which enables both scales and shifts to be learnable. We also propose Outlier-Friendly Feature Knowledge Distillation (OFF) to recover the information lost in extremely low-bit quantization. The proposed OFF can incorporate semantic information and is insensitive to outliers. At the core of OFF is maximizing the mutual information between features in ternarized and floating-point models using cosine similarity. Extensive experiments demonstrate that our TernaryLLM surpasses previous low-bit quantization methods on the standard text generation and zero-shot benchmarks for different LLM families. Specifically, for one of the most powerful open-source models, LLaMA-3, our approach (W1.58A16) outperforms the previous state-of-the-art method (W2A16) by 5.8 in terms of perplexity on C4 and by 8.2% in terms of average accuracy on zero-shot tasks.

cs.LG↗

arXiv2026

Randomized Online Fair Division: High-Probability and Expected Realized Fairness

We study randomized algorithms for the fully online allocation of indivisible goods among $n\ge2$ agents with nonnegative additive valuations. Goods arrive sequentially and must be allocated immediately and irrevocably, with only $n$ known in advance. Since exact ex-ante envy freeness and proportionality are readily achievable, while no positive ex-post approximation is possible for the fairness notions considered here, we study the intermediate notions of high-probability fairness and expected realized fairness. Against a non-adaptive adversary, we give a randomized algorithm for proportionality up to one good (PROP1) whose parameter depends only on $n$ and that preserves exact ex-ante envy-freeness and proportionality. At confidence $1-δ$, its PROP1 guarantee improves on independent uniform allocation (Rand) by a factor of $Ω(\log n)$, uniformly over $δ\in(0,1/2]$. As $n\to\infty$, its expected realized PROP1 factor is at least $\frac{3-\sqrt5}{2}-o(1)$. We also show that the expected realized PROP1 factor of Rand is $(1+o(1))/\log n$, yielding an improvement of at least $\bigl(\frac{3-\sqrt5}{2}-o(1)\bigr)\log n$ for our algorithm. For every randomized online algorithm and every positive approximation factor, the success probability can be made arbitrarily small for envy freeness up to any good (EFX) and at most $\frac{n+1}{2n}$ for envy freeness up to one good (EF1). Consequently, every randomized fully online algorithm has an expected realized EFX guarantee of zero and an expected realized EF1 guarantee of at most $\frac{n+1}{2n}$.

cs.GT↗

arXiv2026

AdaExplore: Failure-Driven Adaptation and Diversity-Preserving Search for Efficient Kernel Generation

Recent large language model (LLM) agents have shown promise in using execution feedback for test-time adaptation. However, robust self-improvement remains far from solved: most approaches still treat each problem instance independently, without accumulating reusable knowledge. This limitation is particularly pronounced in domain-specific languages such as Triton, which are underrepresented in LLM pretraining data. Their strict constraints and non-linear optimization landscape further make naive generation and local refinement unreliable. We propose AdaExplore, an agent framework that enables self-improvement via accumulated execution feedback for performance-critical kernel code generation through two complementary stages: failure-driven adaptation and diversity-preserving search, jointly improving correctness and optimization performance without additional fine-tuning or external knowledge. In the adaptation stage, the agent synthesizes tasks and converts recurring failures into a reusable memory of validity rules, helping subsequent generations remain within the feasible set. In the search stage, the agent organizes candidate kernels as a tree and alternates between small local refinements and larger structural regeneration, allowing it to explore the optimization landscape beyond local optima. Experiments on kernel runtime optimization benchmarks validate these gains: AdaExplore achieves 3.11x and 1.62x speedups on KernelBench Level-2 and Level-3, respectively, within 100 steps, and continues to improve with additional computation.

cs.CL↗

arXiv2026

A Recursive Module-Coupling Algorithm for Computing Low-Energy Eigenstates

Finding the eigenstates of a many-body Hamiltonian is a fundamental challenge in physics and computational science. Since the search space grows exponentially with system size, numerous classical and quantum algorithms have been developed to address this problem. A practical strategy is to identify a physics-informed low-dimensional subspace that effectively accommodates the low-lying eigenstates, thereby reducing the computational complexity. In this paper, we propose a recursive module-coupling algorithm, which iteratively treats a system as a composition of locally-coupled smaller modules, with low-energy subspace estimated successively according to the same recursive structure. Unlike the density matrix renormalization group (DMRG) approach that optimizes a global matrix product state through repeated local sweeps and obtains excited states sequentially, our algorithm constructs a physically tailored variational basis from module eigenstates and obtains several low-energy states on an equal footing, leading to substantial speedups if targeting moderate accuracy. Our proposed method further leads naturally to a recursive quantum variational algorithm, providing a systematic and modular circuit-construction framework compatible with contemporary gate-based quantum architectures. At each recursive level, block encoders are trained to map logical basis states onto the retained physical subspace, within which a variational circuit is subsequently optimized. Such a quantum-circuit implementation provides not only a quantum multistate eigensolver, but also a systematic prescription for hierarchically constructing quantum state-preparation circuits. Classical simulations demonstrate the accuracy and efficiency of the proposed method, whereas experiments on IBM quantum processors show that eigenstate preparation with reasonable fidelities is achievable even in the current NISQ era.

quant-ph↗

arXiv2026

DLM-One: Diffusion Language Models for One-Step Sequence Generation

This paper introduces DLM-One, a score-distillation-based framework for one-step sequence generation with continuous diffusion language models (DLMs). DLM-One eliminates iterative refinement by aligning the scores of a student model's outputs with the score function of a pretrained teacher DLM in the forward-diffused noisy space. We demonstrate that our framework is architecture-agnostic and robust across diverse continuous manifolds, including standard token embedding spaces and logit simplex spaces. Through experiments on multiple representative DLMs, we show that DLM-One achieves up to $\sim$2000$\times$ speedup in sampling steps and $\sim$500$\times$ in wall-clock time, while maintaining competitive performance on benchmark text generation tasks. We further analyze failure modes in language-domain diffusion distillation and propose an adversarially-regularized two-stage training scheme to prevent student degeneration. Our findings position one-step score distillation as a viable path for the efficient deployment of continuous diffusion models operating in continuous space for natural language processing.

cs.CL↗

arXiv2026

Resource-Efficient Bio-Molecular Docking on a NISQ-era Digital Quantum Computer

Molecular docking is a vital computational task in drug discovery, wherein the objective is to efficiently identify optimal binding poses between a ligand and a target receptor protein. Due to the combinatorial explosion of possible binding configurations, docking of large and flexible molecules remains a computationally intensive problem, especially at scale. Early studies have revealed that the molecular docking can be re-cast as a maximum vertex-weighted clique problem (MVWCP) problem on a compatibility graph to be solved classically. In this work, we proposed a hybrid quantum-classical approach for molecular docking leveraging the MVWCP formalism with a variational full-basis encoding (FBE) strategy, which enables efficient encoding of classical binary variables with Bloch sphere vectors. We further prove that a global minimizer of the FBE objective can always be chosen to be a pure product state, thereby providing a rigorous justification for its optimization using a unitary variational circuit. The molecular docking problem is first mapped to a cost Hamiltonian that is minimized within a variational framework, optimized via a randomized imaginary time evolution (ITE)-inspired warm start, and gradient-based techniques. Finally, we also executed the circuit on an IBM quantum computer, underlying the feasibility and of quantum-assisted optimization for structure-based drug design and point towards the broader utility of advanced encoding techniques in quantum optimization for computational biology.

quant-ph↗

arXiv2026

CAKE: Compiler-Agent Co-Design for Frontier Kernel Evolution

GPU kernel agents and GPU programming languages have advanced separately, leaving expert kernels difficult to reproduce. Agents usually treat the compiler as a fixed black box and receive only errors, correctness outcomes, and timing, while existing DSLs either hide critical scheduling decisions or expose them through difficult layout abstractions. We present CAKE, a compiler-agent co-design in which agents author CAKE IR, a typed, hardware-explicit schedule representation. CAKE exposes warp roles, memory movement, synchronization, and pipelines while supporting verification, cost modeling, and localized diagnostics. The harness itself evolves: recurring failures become verifier rules, IR primitives, model calibrations, and reusable optimization tactics. In matched implementation-hidden Flash-KMeans clean starts on B200, the best CAKE IR candidate at an 80-million-token budget runs at 1.144x the tuned FlashML baseline, compared with 0.928x for direct CUDA/PTX. Beyond this benchmark, agent-generated Kimi Delta Attention achieves a 2.05x geometric-mean speedup over official FlashKDA and passes end-to-end serving validation. Dispatcher-backed KNN and KMeans improve performance by 1.42x to 2.12x across more than 400 shapes, and four kernel changes are available as upstream PRs. CAKE targets NVIDIA GPUs from Ampere through Blackwell and separates single-shape evolution from library generalization and dispatch.

cs.LG↗

arXiv2026

XGrammar-2: Dynamic and Efficient Structured Generation Engine for Agentic LLMs

Modern LLM agents increasingly rely on dynamic structured generation, such as tool calling and response protocols. Unlike traditional structured generation with static structures, these workloads vary both across requests and within a request, posing new challenges to existing engines. We present XGrammar-2, a structured generation engine for dynamic agentic workloads. Our design is based on two key ideas: first-class support for tag-triggered structure switching, and fine-grained reuse across requests with different output structures. Concretely, XGrammar-2 introduces TagDispatch for dynamic structural dispatching and Cross-Grammar Cache for substructure-level cache reuse across grammars. It further improves efficiency with an Earley-based adaptive token mask cache, just-in-time compilation, and repetition state compression. Experiments show that XGrammar-2 achieves over 6x faster compilation than prior structured generation engines, and incurs near-zero end-to-end overhead in modern LLM serving systems.

cs.AI↗

arXiv2026

Manipulation of Topological Corner States via Subchiral Symmetry

Higher-order topological phases provide robust corner modes, but their use requires controllable creation, isolation, and transfer of individual modes and their superpositions. Here we demonstrate, using the two-dimensional Benalcazar-Bernevig-Hughes model as an example, that subchiral symmetry provides a general control principle for manipulating topological corner modes. The conventional chiral symmetry decomposes into four subchiral symmetries, each associated with one zero-energy corner mode. By selectively breaking these subsymmetries with controlled intercell hoppings, we reduce the fourfold corner-state manifold step by step to single isolated modes. We further design adiabatic protocols that transfer either a single corner state or a superposition of two corner states between selected corners, while preserving the relative phase in the latter case. Both numerical simulations and IBM quantum-processor implementations show that the proposed protocols can be executed with high fidelity, establishing subchiral symmetry as a route to programmable higher-order topological state manipulation.

quant-ph↗

arXiv2026

PPDM: Pixel Puzzling Diffusion Model for Speed and Memory Efficient Volumetric Medical Image Translation

Diffusion models have demonstrated superior fidelity for medical image-to-image translation, but their extension to high-resolution 3D volumes is severely constrained by prohibitive computational cost and GPU memory requirements. Existing memory-efficient strategies often compromise global volumetric consistency or fine anatomical detail. In this work, we propose the Pixel Puzzling Diffusion Model (PPDM), a simple and effective framework for memory- and speed-efficient 3D medical image translation. PPDM introduces a reversible pixel puzzle-unpuzzle operator that trades spatial resolution for channel dimensionality, substantially reducing activation memory while preserving global context. To further improve efficiency and stability, we adopt a direct bridge diffusion formulation that starts from the conditional input rather than pure noise, enabling the model to focus on task-relevant residuals. In addition, a puzzle-gradient loss is incorporated to enforce spatial coherence and suppress grid-like artifacts introduced by spatial rearrangement. We evaluate PPDM on multiple challenging 3D medical image translation tasks, including low-count PET denoising, joint PET denoising and attenuation correction, and cross-modal MRI translation. Across all tasks, PPDM consistently matches or outperforms full 3D diffusion models while reducing training GPU memory usage by up to an order of magnitude and significantly accelerating inference, and it outperforms existing memory-efficient diffusion approaches based on latent compression or frequency decomposition. These results demonstrate that PPDM provides a practical and scalable solution for high-fidelity 3D diffusion-based medical image translation under limited computational resources.

cs.CV↗

arXiv2026

Competitive Analysis for Online Fair Division under Multiple Fairness Notions

We study the online fair division of indivisible items with additive utilities, where items arrive sequentially and must be irrevocably allocated upon arrival. Considering various fairness notions, we focus on designing online algorithms that produce fair or approximately fair allocations for any instance. We measure algorithm performance using the competitive ratio, defined as the worst-case ratio between the fairness guarantee achieved by the online algorithm and that of an optimal offline allocation with full knowledge of future arrivals. We examine a broad spectrum of models, including the allocation of goods or chores, normalized versus non-normalized utilities, and identical versus general utility functions. We address the majority of the unresolved cases by providing online algorithms or proving the limits of the competitive ratio achievable by online algorithms. In most cases, the algorithms are shown to be optimal.

cs.GT↗

arXiv2026

MPK: A Compiler and Runtime for Mega-Kernelizing Tensor Programs

We introduce Mirage Persistent Kernel (MPK), the first compiler and runtime system that automatically transforms multi-GPU model inference into a single high-performance mega-kernel. MPK introduces an SM-level graph representation that captures data dependencies at the granularity of individual streaming multiprocessors (SMs), enabling cross-operator software pipelining, \rev{fine-grained overlap of computation and communication, and other optimizations that are infeasible under the conventional kernel-per-operator execution model}. The MPK compiler lowers tensor programs into optimized SM-level task graphs and generates fast CUDA implementations for each task, while the MPK in-kernel parallel runtime executes these tasks within a single persistent mega-kernel using decentralized scheduling across SMs. Together, these components provide end-to-end kernel fusion with minimal developer effort, while preserving the flexibility of existing programming models. Our evaluation shows that MPK significantly outperforms existing kernel-per-operator LLM serving systems, achieving up to 1.7$\times$ lower end-to-end inference latency and pushing LLM inference performance close to the limits of the underlying hardware. MPK is publicly available at https://github.com/mirage-project/mirage.

cs.DC↗

arXiv2026

Less Is More: Training-Free Acceleration Framework of 3D Diffusion Models for Low-Count PET Denoising via Global-Local Trajectory Reduction

Accurate quantification and uptake measurement in PET are critical for assessing disease progression and supporting clinical decision-making. While high-count PET provides reliable image quality, the associated radiation dose and prolonged acquisition remain significant clinical concerns, motivating the adoption of low-count protocols. Diffusion-model-based methods have demonstrated strong potential for restoring low-count PET to near high-count quality, but their iterative sampling procedure becomes prohibitively expensive when applied to high-resolution 3D PET volumes, introducing substantial inference latency that limits practical clinical deployment. To address these challenges, we propose a training-free Global-Local Skipping Strategy that accelerates diffusion model-based 3D PET denoising while simultaneously improving reconstruction quality. The proposed method is plug-and-play and directly applicable to pre-trained diffusion models without retraining or architectural modification. Specifically, we introduce: (i) a global denoising step skipping strategy that initializes the reverse diffusion process from an intermediate denoising step using a noise-consistent transformation of the low-count input, substantially reducing the number of required denoising steps; and (ii) a local feature reuse shortcut that reuses slowly-varying high-level U-Net features across neighboring denoising steps, further reducing per-step computation while preserving image fidelity. We evaluate the proposed approach on multiple PET tracers from in-house and public datasets, including 18F-FDG PET, 68Ga-DOTATATE PET, and 18F-PSMA PET, demonstrating consistent acceleration of over an order of magnitude alongside improved or comparable reconstruction performance relative to the full-step baseline. Blinded reader studies further confirm enhanced clinical confidence and perceived diagnostic quality.

cs.CV↗

arXiv2026

Simulating Condensed Matter Physics on Quantum Hardware

Quantum hardware platforms are getting increasingly sophisticated in their ability to simulate condensed matter, including but not limited to strongly-correlated, topological, and non-equilibrium phenomena. This review surveys recent progress in quantum-hardware-based simulations of condensed matter, primarily emphasizing gate-based digital quantum computer simulation, with analog experiments discussed as complementary benchmarks. We first review major hardware platforms, including superconducting qubits, trapped-ions, ultracold atoms, Rydberg arrays, photonic systems, and moire quantum materials. We then introduce the basic ingredients of digital quantum simulation. Building on this foundation, we discuss representative applications to condensed-matter physics, spanning ground-state problems, strongly correlated matter, topological phases, non-equilibrium dynamics, open-system physics, and high-energy-physics-inspired simulations. Finally, we summarize key methodological tools used in state-of-the-art quantum-simulation workflows. We emphasize that present noisy quantum simulations serve not only as near-term demonstrations, but also as prototypes for the encodings, diagnostic protocols and error-control strategies required for future fault-tolerant quantum simulation.

cond-mat.str-el↗

arXiv2026

PithTrain: A Compact and Agent-Native MoE Training System

Mixture-of-Experts (MoE) has become the dominant architecture for frontier language models. To meet this demand, production frameworks have built optimized MoE training stacks over years of engineering effort. Yet evolving these stacks for new architectures and system optimizations remains expensive. With the rise of AI coding agents, they could automate parts of training-framework development and accelerate this evolution. But applying them to these existing frameworks carries hidden costs, invisible to today's throughput-only evaluations. We name this missing dimension agent-task efficiency (ATE): the cost of using coding agents to understand, operate, and extend a framework. Grounded in four agent-native design principles, we build PithTrain, a compact, agent-native MoE training framework. We further introduce ATE-Bench, covering real-world training-framework tasks. Our evaluation shows PithTrain matches the throughput of production frameworks, and on ATE-Bench, PithTrain enables higher agent-task efficiency, with up to 62% fewer Agent Turns and 64% less Active GPU Time.

cs.LG↗