arXiv subjects

Chonghan Liu

Publications and source records attributed to Chonghan Liu.

At least 19 recordsLinked to original sources

arXiv2026

CAPF: Guiding Search-Agent Rollouts with Credit-Attenuated Privileged Feedback

Recent LLM search agents use reinforcement learning with verifiable rewards (RLVR) to learn search-augmented reasoning from outcome rewards. On hard problems, these agents rarely sample end-to-end successful rollouts, leaving outcome-only RLVR with few positive-reward trajectories. We argue that improving learning on such problems requires additional guidance during training, and RLVR already contains verifier-side information that can provide it. This information can identify errors or omissions in the agent's submitted answer and guide revision within the rollout. We propose a training-time mechanism called \textbf{Credit-Attenuated Privileged Feedback} (CAPF), which makes this verifier-side information available through a Privileged Feedback call during training. CAPF lets the policy revise zero-reward attempts into positive-reward repair trajectories and attenuates credit for the feedback call and earlier actions to accommodate deployment without this call. Empirical research demonstrates that CAPF improves Qwen3-4B's average exact-match score from 44.7% under outcome-only RLVR to 48.5% on seven open-domain QA benchmarks.

cs.AI↗

arXiv2026

Rethinking LLM Ensembling from the Perspective of Mixture Models

Model ensembling is a well-established technique for improving the performance of machine learning models. Conventionally, this involves averaging the output distributions of multiple models and selecting the most probable label. This idea has been naturally extended to large language models (LLMs), yielding improved performance but incurring substantial computational cost. This inefficiency stems from directly applying conventional ensemble implementation to LLMs, which require a separate forward pass for each model to explicitly compute the ensemble distribution. In this paper, we propose the Mixture-model-like Ensemble (ME). By reinterpreting the ensemble as a mixture model, ME stochastically selects a single model at each step to generate the next token, thereby avoiding the need to explicitly compute the full ensemble distribution. ME is mathematically equivalent to sampling from the ensemble distribution, but requires invoking only one model, making it 1.78x-2.68x faster than conventional ensembling. Furthermore, this perspective connects LLM ensembling and token-level routing methods, suggesting that LLM ensembling is a special case of routing methods. Our findings open new avenues for efficient LLM ensembling and motivate further exploration of token-level routing strategies for LLMs. Our code is available at https://github.com/Kamichanw/Mixture-model-like-Ensemble.

cs.LG↗

arXiv2026

VEPO: Variable Entropy Policy Optimization for Low-Resource Language Foundation Models

Large language models frequently exhibit suboptimal performance on low resource languages, primarily due to inefficient subword segmentation and systemic training data imbalances. In this paper, we propose Variable Entropy Policy Optimization (VEPO), which leverages Reinforcement Learning with Verifiable Rewards to incorporate deterministic structural constraints into the policy alignment process. This framework ensures prescribed sequence length, robust format consistency, and rigorous linguistic well formedness, all enforced during training. Central to our approach is a variable entropy mechanism that enables the model to dynamically calibrate the equilibrium between literal fidelity and semantic naturalness by modulating the exploration exploitation manifold. By integrating entropy tempered advantage estimation with asymmetric clipping, VEPO sustains robust exploration while mitigating policy collapse. Empirical evaluations across 90 FLORES-200, COMET-22, chrF directions demonstrate that VEPO yields substantial improvements in both tokenization efficiency and translation quality, bridging the performance gap for underrepresented languages.

cs.CL↗

arXiv2026

Modality Gap-Driven Subspace Alignment Training Paradigm For Multimodal Large Language Models

Despite the success of multimodal contrastive learning in aligning visual and linguistic representations, a persistent geometric anomaly, the Modality Gap, remains: embeddings of distinct modalities expressing identical semantics occupy systematically offset regions. Prior approaches to bridge this gap are largely limited by oversimplified isotropic assumptions, hindering their application in large-scale scenarios. In this paper, we address these limitations by precisely characterizing the geometric shape of the modality gap and leveraging it for efficient model scaling. First, we propose the Fixed-frame Modality Gap Theory, which decomposes the modality gap within a frozen reference frame into stable biases and anisotropic residuals. Guided by this precise modeling, we introduce ReAlign, a training-free modality alignment strategy. Utilizing statistics from massive unpaired data, ReAlign aligns text representation into the image representation distribution via a three-step process comprising Anchor, Trace, and Centroid Alignment, thereby explicitly rectifying geometric misalignment. Building on ReAlign, we propose ReVision, a scalable training paradigm for Multimodal Large Language Models~(MLLMs). ReVision integrates ReAlign into the pretraining stage, enabling the model to learn the distribution of visual representations from unpaired text before visual instruction tuning, without the need for large-scale, high-quality image-text pairs. Our framework demonstrates that statistically aligned unpaired data can effectively substitute for expensive image-text pairs, offering a robust path for the efficient scaling of MLLMs.

cs.CV↗

arXiv2026

Flatter Tokens are More Valuable for Speculative Draft Model Training

Speculative Decoding (SD) is a key technique for accelerating Large Language Model (LLM) inference, but it typically requires training a draft model on a large dataset. We approach this problem from a data-centric perspective, finding that not all training samples contribute equally to the SD acceptance rate. Specifically, our theoretical analysis and empirical validation reveals that tokens inducing flatter predictive distributions from the target model are more valuable than those yielding sharply peaked distributions. Based on this insight, we propose flatness, a new metric to quantify this property, and develop the Sample-level-flatness-based Dataset Distillation (SFDD) approach, which filters the training data to retain only the most valuable samples. Experiments on the EAGLE framework demonstrate that SFDD can achieve over 2$\times$ training speedup using only 50% of the data, while keeping the final model's inference speedup within 4% of the full-dataset baseline. This work introduces an effective, data-centric approach that substantially improves the training efficiency for Speculative Decoding. Our code is available at https://github.com/fjm9933/Flatness.

cs.CL↗

arXiv2026

Radiation Resistance of Ge-doped Multi-Mode Fiber for Optical Links in Collider Experiments

The applications of optical links in collider experiments provide the advantage of high-speed data transmission with low mass fibers over distances of a few hundred meters. Ge-doped multi-mode fibers are evaluated for radiation tolerance in ionizing doses of Co-60 gamma rays. The Radiation-Induced Attenuation (RIA) varies significantly depending on doping substances and fabrication technologies. A type of telecom-grade fiber has demonstrated an RIA of 0.05 dB/m under a total ionizing dose of 300 kGy(SiO2). The dependence on dose rate is compared in the range between 5 Gy/hr and 1.4 kGy/hr, and the annealing recovery is observed after the Co-60 source is shielded. The temperature dependence is investigated across a range of -15 oC to room temperature. At cold temperatures, stagnant annealing leads to a substantially higher RIA during irradiation. The recovery of radiation-induced defects is typically within a few hours, resulting in similar RIA levels regardless of the dose rate and temperature during exposure. Ge-doped fibers of chosen fabrication methods are capable of enduring high ionizing doses for use in high-energy physics experiments.

hep-ex↗

arXiv2025

HI-TransPA: Hearing Impairments Translation Personal Assistant

Hearing-impaired individuals often face significant barriers in daily communication due to the inherent challenges of producing clear speech. To address this, we introduce the Omni-Model paradigm into assistive technology and present HI-TransPA, an instruction-driven audio-visual personal assistant. The model fuses indistinct speech with lip dynamics, enabling both translation and dialogue within a single multimodal framework. To address the distinctive pronunciation patterns of hearing-impaired speech and the limited adaptability of existing models, we develop a multimodal preprocessing and curation pipeline that detects facial landmarks, stabilizes the lip region, and quantitatively evaluates sample quality. These quality scores guide a curriculum learning strategy that first trains on clean, high-confidence samples and progressively incorporates harder cases to strengthen model robustness. Architecturally, we employs a novel unified 3D-Resampler to efficiently encode the lip dynamics, which is critical for accurate interpretation. Experiments on purpose-built HI-Dialogue dataset show that HI-TransPA achieves state-of-the-art performance in both literal accuracy and semantic fidelity. Our work establishes a foundation for applying Omni-Models to assistive communication technology, providing an end-to-end modeling framework and essential processing tools for future research.

cs.CL↗

arXiv2025

Does RLVR Extend Reasoning Boundaries? Investigating Capability Expansion in Vision-Language Models

Recent studies posit that Reinforcement Learning with Verifiable Rewards (RLVR) primarily amplifies behaviors inherent to the pre-training distribution rather than inducing new capabilities, but these insights are predominantly limited to language-only domains, leaving the dynamics of visual-centric spatial reasoning under-explored. To examine the impact of RLVR on the capability boundaries of Vision-Language Models (VLMs), we introduce \textbf{Ariadne}, a controlled framework based on synthetic maze navigation where the reasoning difficulty is precisely regulated by path length and the number of turns. We demonstrate that applying RLVR extends the spatial reasoning boundary, achieving success on problems where the base policy VLM consistently attains $0\%$ accuracy despite increasing pass@k sampling budgets, indicating that the optimized policy successfully navigates search spaces that were effectively unreachable by the base distribution. Furthermore, despite being trained exclusively on synthetic mazes, we evaluate the model on two real-world navigation benchmarks (MapBench and ReasonMap) in a zero-shot setting. The observed improvements in these out-of-domain tasks suggest genuine spatial reasoning capability expansion rather than mere sampling efficiency.

cs.AI↗

arXiv2025

d$^2$Cache: Accelerating Diffusion-Based LLMs via Dual Adaptive Caching

Diffusion-based large language models (dLLMs), despite their promising performance, still suffer from inferior inference efficiency. This is because dLLMs rely on bidirectional attention and cannot directly benefit from the standard key-value (KV) cache as autoregressive models (ARMs) do. To tackle this issue, we introduce \textit{Dual aDaptive Cache} (d$^2$Cache), which is a training-free approximate KV cache framework for accelerating dLLM inference. d$^2$Cache features a two-stage fine-grained selection strategy to identify tokens and adaptively update their KV states at each decoding step, while caching the KV states of the remaining tokens for reuse. Furthermore, d$^2$Cache naturally offers a more reliable decoding alternative, which can enable quasi left-to-right generation and mitigate premature overconfidence in tokens at the end of the sequence. Extensive experimental results on two representative dLLMs (\ie, LLaDA and Dream) demonstrate that d$^2$Cache not only achieves substantial inference speedups, but also yields consistent improvements in generation quality. The code is available at https://github.com/Kamichanw/d2Cache.

cs.CL↗

arXiv2025

MIRAGE: A Multi-modal Benchmark for Spatial Perception, Reasoning, and Intelligence

Spatial perception and reasoning are core components of human cognition, encompassing object recognition, spatial relational understanding, and dynamic reasoning. Despite progress in computer vision, existing benchmarks reveal significant gaps in models' abilities to accurately recognize object attributes and reason about spatial relationships, both essential for dynamic reasoning. To address these limitations, we propose MIRAGE, a multi-modal benchmark designed to evaluate models' capabilities in Counting (object attribute recognition), Relation (spatial relational reasoning), and Counting with Relation. Through diverse and complex scenarios requiring fine-grained recognition and reasoning, MIRAGE highlights critical limitations in state-of-the-art models, underscoring the need for improved representations and reasoning frameworks. By targeting these foundational abilities, MIRAGE provides a pathway toward spatiotemporal reasoning in future research.

cs.CV↗

arXiv2025

Characteristics of Ge-doped Multi-Mode Fibers in Total Ionizing Dose

Purpose: The fiber optical links in 850 nm band with Ge-doped multi-mode (MM) fibers are well developed for data transmission at 10 Gbps and higher. The applications in nuclear environments require radiation resistance. The characteristics of Ge-doped MM fibers are investigated for Radiation Induced Attenuation (RIA) in Total Ionizing Dose (TID). Methods: Commercial samples of Ge-doped MM fibers were irradiated in Go-60 gamma rays at dose rates of 5 to 1.4k Gy(SiO2)/hr. The fiber samples were packaged in water tanks maintained at constant temperatures in the range of -15 to 45 degC. The optical power transmitted through the fibers were recorded in irradiation, and in annealing when the source was shielded. The measurements of RIA in time are analyzed for dose rate and temperature dependences. Results: Ge-doped fiber samples of OM2 to OM4 grades were investigated for attenuation of optical power in radiation ionizing dose. Depending on the fabrication technology, two of the fiber types show radiation resistance with the RIAs of 0.2 dB/m and 0.05 dB/m, respectively, for the TID of 300 kGy(SiO2). At low dose rate of 5 Gy/hr, the RIA increases steadily and the annealing of low density ionizing defects does not cause notable deviation. At 1.4 kGy/hr the accumulated defects result to twice higher RIA during irradiation, and is worsen to a factor three in cold temperature. However, once the source is shielded the recovery is effective in a few hours. Conclusion: The telecom products of 850 nm Ge-doped MM fibers provide high speed communication in distances of a few hundred meters. The industrial fabrication methods provide fibers that can endure radiation ionizing dose for applications in nuclear instrumentation.

physics.med-ph↗

arXiv2025

AdaptiveStep: Automatically Dividing Reasoning Step through Model Confidence

Current approaches for training Process Reward Models (PRMs) often involve breaking down responses into multiple reasoning steps using rule-based techniques, such as using predefined placeholder tokens or setting the reasoning step's length into a fixed size. These approaches overlook the fact that specific words do not typically mark true decision points in a text. To address this, we propose AdaptiveStep, a method that divides reasoning steps based on the model's confidence in predicting the next word. This division method provides more decision-making information at each step, enhancing downstream tasks, such as reward model learning. Moreover, our method does not require manual annotation. We demonstrate its effectiveness through experiments with AdaptiveStep-trained PRMs in mathematical reasoning and code generation tasks. Experimental results indicate that the outcome PRM achieves state-of-the-art Best-of-N performance, surpassing greedy search strategy with token-level value-guided decoding, while also reducing construction costs by over 30% compared to existing open-source PRMs. In addition, we provide a thorough analysis and case study on the PRM's performance, transferability, and generalization capabilities.

cs.AI↗

arXiv2024

ETROC1: The First Full Chain Precision Timing Prototype ASIC for CMS MTD Endcap Timing Layer Upgrade

We present the design and characterization of the first full chain precision timing prototype ASIC, named ETL Readout Chip version 1 (ETROC1) for the CMS MTD endcap timing layer (ETL) upgrade. The ETL utilizes Low Gain Avalanche Diode (LGAD) sensors to detect charged particles, with the goal to achieve a time resolution of 40 - 50 ps per hit, and 30 - 40 ps per track with hits from two detector layers. The ETROC1 is composed of a 5 x 5 pixel array and peripheral circuits. The pixel array includes a 4 x 4 active pixel array with an H-tree shaped network delivering clock and charge injection signals. Each active pixel is composed of various components, including a bump pad, a charge injection circuit, a pre-amplifier, a discriminator, a digital-to-analog converter, and a time-to-digital converter. These components play essential roles as the front-end link in processing LGAD signals and measuring timing-related information. The peripheral circuits provide clock signals and readout functionalities. The size of the ETROC1 chip is 7 mm x 9 mm. ETROC1 has been fabricated in a 65 nm CMOS process, and extensively tested under stimuli of charge injection, infrared laser, and proton beam. The time resolution of bump-bonded ETROC1 + LGAD chipsets reaches 42 - 46 ps per hit in the beam test.

physics.ins-det↗

arXiv2024

The 120Gbps VCSEL Array Based Optical Transmitter (ATx) Development for the High-Luminosity LHC (HL-LHC) Experiments

The integration of a Verticle Cavity Surface-Emitting Laser (VCSEL) array and a driving Application-Specific Integrated Circuit (ASIC) in a custom optical array transmitter module (ATx) for operation in the detector front-end is constructed, assembled and tested. The ATx provides 12 parallel channels with each channel operating at 10 Gbps. The optical transmitter eye diagram passes the eye mask and the bit-error rate (BER) less than 1E-12 transmission is achieved at 10 Gbps/ch. The overall insertion loss including the radiation induced attenuation is sufficiently low to meet the proposed link budget requirement.

physics.ins-det↗

arXiv2024

Optical Data Transmission ASICs for the High-Luminosity LHC (HL-LHC) Experiments

We present the design and test results of two optical data transmission ASICs for the High-Luminosity LHC (HL-LHC) experiments. These ASICs include a two-channel serializer (LOCs2) and a single-channel Vertical Cavity Surface Emitting Laser (VCSEL) driver (LOCld1V2). Both ASICs are fabricated in a commercial 0.25-um Silicon-on-Sapphire (SoS) CMOS technology and operate at a data rate up to 8 Gbps per channel. The power consumption of LOCs2 and LOCld1V2 are 1.25 W and 0.27 W at 8-Gbps data rate, respectively. LOCld1V2 has been verified meeting the radiation-tolerance requirements for HL-LHC experiments.

physics.ins-det↗

arXiv2024

The Design of a High Speed Low Power Phase Locked Loop

The upgrade of the ATLAS Liquid Argon Calorimeter readout system calls for the development of radiation tolerant, high speed and low power serializer ASIC. We have designed a phase locked loop using a commercial 0.25 um Silicon-on-Sapphire (SoS) CMOS technology. Post-layout simulation indicates that tuning range is 3.79-5.01 GHz and power consumption is 104 mW. The PLL has been submitted for fabrication. The design and simulation results are presented.

physics.ins-det↗

arXiv2024

The Clock Distribution System for the ATLAS Liquid Argon Calorimeter Phase-I Upgrade Demonstrator

A prototype Liquid-argon Trigger Digitizer Board (LTDB), called the LTDB Demonstrator, has been developed to demonstrate the functions of the ATLAS Liquid Argon Calorimeter Phase-I trigger electronics upgrade. Forty Analog-to-Digital converters and four FPGAs with embedded multi-gigabit-transceivers on each Demonstrator need high quality clocks. A clock distribution system based on commercial components has been developed for the Demonstrator. The design of the clock distribution system is presented. The performance of the clock distribution system has been evaluated. The components used in the clock distribution system have been qualified to meet radiation tolerance requirements of the Demonstrator.

physics.ins-det↗

arXiv2024

The VCSEL-based Array Optical Transmitter (ATx) Development Towards 120-Gbps Link for Collider Detector: Development Update

A compact radiation-tolerant array optical transmitter module (ATx) is developed to provide data transmission up to 10Gbps per channel with 12 parallel channels for collider detector applications. The ATx integrates a Vertical Cavity Surface-Emitting Laser (VCSEL) array and driver circuitry for electrical to optical conversion, an edge warp substrate for the electrical interface and a micro-lens array for the optical interface. This paper reports the continuing development of the ATx custom package. A simple, high-accuracy and reliable active-alignment method for the optical coupling is introduced. The radiation-resistance of the optoelectronic components is evaluated and the inclusion of a custom-designed array driver is discussed.

physics.ins-det↗