arXiv subjects

Tianqi Gao

Publications and source records attributed to Tianqi Gao.

13 recordsLinked to original sources

arXiv2026

AUSO: Action-Level Unified Skill Optimization from Internalization to Utilization

Skills play different roles as an agent's policy evolves: they should first provide learnable knowledge, then support capability formation, and finally be invoked only when they improve individual decisions. Existing methods rarely model this lifecycle. They either keep skills outside the model, fully internalize them, or select among internalization and utilization objectives through noisy task-level success rates. Such designs fragment training and assign uniform importance to actions within the same trajectory, even though skill guidance may help some decisions while distracting others. To solve these problems, we introduce AUSO (Action-level Unified Skill Optimization), which unifies skill learning and skill use through a progressive, action-aware optimization process. At the beginning of training, AUSO jointly learns from teacher guidance and environmental outcomes, enabling the policy to acquire foundational skills without losing task-oriented feedback. It subsequently emphasizes outcome-based policy optimization to consolidate autonomous problem-solving ability. As the policy matures, AUSO evaluates each sampled action under both skill-conditioned and skill-free contexts. The resulting action-level information signal is coupled with the trajectory outcome advantage, allowing beneficial skill-sensitive actions to receive stronger updates and harmful ones to be suppressed. Therefore, skills gradually transition from an external source of supervision into decision knowledge whose utilization is adapted to its action-level benefit, while reinforcement learning remains the shared backbone across all stages. Experiments on ALFWorld, WebShop, and SearchQA show that AUSO consistently improves agent performance and out-of-distribution generalization over competitive baselines.

cs.AI↗

arXiv2026

Beyond Item Order: Temporal Gap Tokenization for Generative Recommendation with Semantic IDs

Semantic-ID-based generative recommendation has recently emerged as a scalable paradigm for sequential recommendation, where each item is represented by a compact sequence of discrete codes and next-item prediction is formulated as code generation. Existing methods, however, typically construct user histories as sequences of static item identifiers, leaving the elapsed time between consecutive interactions outside the generative input. This temporal blindness is problematic because inter-interaction gaps provide useful cues about interest continuity and preference drift. In this paper, we propose ChronoSID, a lightweight temporal augmentation framework for semantic-ID-based generative recommendation. ChronoSID injects temporal signals into the standard three-stage semantic-ID pipeline from two complementary perspectives. First, we introduce Time-Aware Field-Aware Masked Auto-Encoding (TA-FAMAE), which regularizes item representation learning with an auxiliary time-gap prediction objective. Second, we discretize historical interaction intervals into fixed log-scale gap tokens and interleave them with semantic ID tuples as the encoder input of the sequence-to sequence generator. This design preserves the compact SID generation paradigm while enabling the model to capture time-aware transition patterns. Experiments on Amazon review benchmarks show that ChronoSID consistently improves over ReSID and other competitive generative recommendation baselines. Ablation studies further verify the contribution of both temporal components, and diagnostic analyses show clearer gains under long-gap scenarios where user interests are more likely to drift.

cs.IR↗

arXiv2026

A Reproducible Benchmark and Evidence-Retrieval Software Framework for Silicon Detector R&D Literature

Silicon pixel detector R&D depends on a large and rapidly growing technical literature, including beam-test and irradiation studies, performance measurements, simulation, and design reports. Locating the supporting evidence passage for a measurement, operating condition, or design decision is therefore a computing and data-science challenge for detector-development workflows. General-purpose language models are insufficient unless grounded in traceable primary sources, particularly in a domain with specialised terminology, configuration-dependent measurements, and rapidly evolving experimental results. We address this with a reproducible, general-purpose framework for evidence-grounded retrieval over technical literature, using silicon pixel detector R&D as a demanding validation domain. The framework combines sparse lexical retrieval, dense semantic retrieval, and hybrid reciprocal-rank fusion, with an optional graph-guided exploration layer and grounded, abstention-aware response generation. The accompanying benchmark provides manually curated chunk-level evidence annotations, source-level diagnostics, semantic relevance checks, and negative-query abstention tests over two detector query sets. We evaluate six retrieval configurations across 378 source documents and 8,442 indexed chunks. Hybrid sparse-dense retrieval gives the strongest strict evidence recovery, achieving Hit@5 of 0.917 on the core benchmark and 0.951 on the curated extension benchmark, while graph-based methods are more effective for literature exploration and source discovery. Graph expansion is therefore best employed as a discovery layer over the hybrid retrieval backbone. The framework provides reusable software for traceable, 1 evidence-grounded knowledge access in silicon detector R&D and high-energy physics instrumentation.

physics.ins-det↗

arXiv2026

ThinkDeception: A Progressive Reinforcement Learning Framework for Interpretable Multimodal Deception Detection

Multimodal deception detection is critical for identifying fraudulent intentions, yet existing approaches predominantly rely on end to end black--box paradigms. These methods suffer from a severe lack of interpretability failing to provide transparent reasoning trajectories and struggling to explicitly capture the subtle, cross modal inconsistencies inherent in deceptive behaviors. To transcend these limitations, we propose ThinkDeception, a novel and interpretable multimodal deception detection framework. As a pioneering effort, it introduces Multimodal Large Language Models (MLLMs) into this domain, transforming deception detection from a traditional binary classification task into an explicit cognitive reasoning process. Facilitated by the first meticulously annotated step--by--step multimodal Chain of Thought (CoT) dataset, we develop a foundational model, ThinkDeception Base, empirically validating the critical role of modal inconsistency in decoding deception. Building upon this foundation, our core innovation lies in proposing Visual-Audio Consistency Group Relative Policy Optimization(VAC--GRPO) equipped with a progressive training strategy. Distinct from standard GRPO, we stratify the training data into four progressive difficulty tiers, guiding the model through a psychologically grounded easy--to--hard cognitive transition. By innovatively coupling this dynamic curriculum scheduler with a multi dimensional, process aware reward mechanism and a reflective learning paradigm, we significantly elevate the model's overall reasoning quality. Extensive experiments on mainstream benchmarks demonstrate that ThinkDeception establishes a new SOTA, significantly outperforming existing methods in both detection accuracy and rationale quality. Ultimately, this work successfully drives the field of deception detection toward interpretable, multimodal cognitive reasoning.

cs.AI↗

arXiv2026

Factorized Latent Reasoning for LLM-based Recommendation

Large language models (LLMs) have recently been adopted for recommendation by framing user preference modeling as a language generation problem. However, existing latent reasoning approaches typically represent user intent with a single latent vector, which struggles to capture the inherently multi-faceted nature of user preferences. We propose Factorized Latent Reasoning (FLR), a novel framework for LLM-based sequential recommendation that decomposes latent reasoning into multiple disentangled preference factors. FLR introduces a lightweight multi-factor attention module that iteratively refines a latent thought representation, where each factor attends to distinct aspects of the user's interaction history. To encourage diversity and specialization, we design orthogonality, attention diversity, and sparsity regularization objectives, and dynamically aggregate factor contributions for the final prediction. We further integrate FLR with an efficient reinforcement learning strategy based on group-relative policy optimization, enabling stable alignment directly in the latent reasoning space. Experiments on multiple benchmarks show that FLR consistently outperforms strong baselines while improving robustness and interpretability.

cs.IR↗

arXiv2026

Discovery of a nonsymmorphic superconductor with spontaneous rotational symmetry breaking and nontrivial zero modes

Topological superconductivity has attracted great interest due to its fundamental significance for realizing Majorana quasiparticles and fault-tolerant quantum computation. Nonsymmorphic superconductors, with symmetry-protected nontrivial electronic structures, offer a promising route to exotic topological superconducting states, yet experimental realizations remain scarce. Here we identify nonsymmorphic compound PtPb4 as a robust platform hosting superconductivity with spontaneous rotational symmetry breaking and nontrivial zero-energy modes. PtPb4 crystallizes in a frustrated Shastry-Sutherland lattice and exhibits nontrivial band topology. By combining in-plane and out-of-plane resistivity measurements, pronounced twofold anisotropy is observed in both the superconducting state and the upper critical field, evidencing spontaneous rotational symmetry breaking. Scanning tunneling microscopy/spectroscopy further reveal twofold-symmetric magnetic vortices, providing direct real-space evidence for the symmetry-broken superconducting state. Notably, a robust zero-energy vortex bound state emerges and persists without spatial splitting over extended distances, consistent with the characteristics expected for Majorana bound state. These findings uncover an exotic superconducting state in PtPb4 and establish a promising platform for exploring topological superconductivity and superconducting quantum devices.

cond-mat.supr-con↗

arXiv2026

Event-Level Voxel Reconstruction in Two-Photon Absorption Scans Using Pixel-Overlap Selection in Timepix3

Two-photon absorption (TPA) enables three-dimensional characterisation of silicon detectors by generating charge carriers within a confined volume around a focused laser spot. In combination with pixelated readout systems, TPA measurements provide access to spatially resolved timing observables relevant for electric field reconstruction. However, the interpretation of TPA data in segmented detectors is non-trivial: a single excitation produces multi-pixel clusters within the intrinsic time resolution of the readout, and in many implementations no external synchronisation between laser pulses and detector data is available. In this work, we present a reconstruction framework for event-level voxelisation of TPA scans using Timepix3, operating on continuous, unsynchronised data. The method introduces a pixel-overlap-based definition of TPA events and a cluster-level timing estimator based on the highest deposited charge within a region of interest. This approach enables blind reconstruction of dwell structure and stable assignment of voxel timing without external triggers. We demonstrate that commonly used alternatives, such as centroid-based selection or earliest-hit timing, introduce systematic spatial biases in clustered events. The proposed framework provides a robust and general method for reconstructing voxel-resolved timing information in segmented detectors, and is directly applicable to TPA-based studies of electric field distributions and charge transport in silicon sensors.

physics.ins-det↗

arXiv2025

A Neutron Sensitive Detector Using 3D-Printed Scintillators

This work reports on the performance of a novel neutron-sensitive scintillating detector fabricated using Fused-Deposition Modelling (FDM) additive manufacturing. FDM is a cost-effective 3D-printing method employing flexible plastic filaments to create custom-shaped components. Scintillating filaments, based on polystyrene doped with \emph{p}-terphenyl and 1,4-bis (5-phenyloxazol-2-yl) benzene, and enriched with $^6$LiF to enable neutron sensitivity were manufactured in house and achieved visible scintillation with a light output of 30$\pm$5~photons per MeV. Printed scintillators were then integrated into a detector system consisting of an image intensified TimePix3 camera, offering high spatial and temporal resolution. The detector performance was compared with Geant4 simulations of the scintillating sensor's response to electrons, gamma-rays, and thermal neutrons. A novel event discrimination algorithm, using the properties of the TimePix3 camera, enabled the separation of neutron signatures from the gamma-ray background.

physics.ins-det↗

arXiv2024

Feasibility study of a novel thermal neutron detection system using event mode camera and LYSO scintillation crystal

The feasibility study of a new technique for thermal neutron detection using a Timepix3 camera (TPX3Cam) with custom-made optical add-ons operated in event-mode data acquisition is presented. The camera has a spatial resolution of ~ 16 um and a temporal resolution of 1.56 ns. Thermal neutrons react with 6 Lithium to produce a pair of 2.73 MeV tritium and 2.05 MeV alpha particles, which in turn interact with a thin layer of LYSO crystal to produce localized scintillation photons. These photons are directed by a pair of lenses to an image intensifier, before being recorded by the TPX3Cam. The results were reconstructed through a custom clustering algorithm utilizing the Time-of-Arrival (ToA) and geometric centre of gravity of the hits. Filtering parameters were found through data analysis to reduce the background of gamma and other charged particles. The efficiency of the converter is 4%, and the overall detection efficiency of the system including the lead shielding and polythene moderator is ~ 0.34%, all converted thermal neutrons can be seen by the TPX3Cam. The experiment used a weak thermal neutron source against a large background, the measured signal-to-noise ratio is 1/67.5. Under such high noise, thermal neutrons were successfully detected and predicted the reduced neutron rate, and matched the simulated rate of the thermal neutrons converted from the source. This result demonstrated the excellent sensitivity of the system.

physics.ins-det↗

arXiv2024

Automatic Extraction and Compensation of P-Bit Device Variations in Large Array Utilizing Boltzmann Machine Training

Probabilistic Bit (P-Bit) device serves as the core hardware for implementing Ising computation. However, the severe intrinsic variations of stochastic P-Bit devices hinder the large-scale expansion of the P-Bit array, significantly limiting the practical usage of Ising computation. In this work, a behavioral model which attributes P-Bit variations to two parameters α and ΔV is proposed. Then the weight compensation method is introduced, which can mitigate α and ΔV of P-Bits device variations by rederiving the weight matrix, enabling them to compute as ideal identical PBits without the need for weights retraining. Accurately extracting the α and ΔV simultaneously from a large P-Bit array which is prerequisite for the weight compensation method is a crucial and challenging task. To solve this obstacle, we present the novel automatic variation extraction algorithm which can extract device variations of each P-Bit in a large array based on Boltzmann machine learning. In order for the accurate extraction of variations from an extendable P-Bit array, an Ising Hamiltonian based on 3D ferromagnetic model is constructed, achieving precise and scalable array variation extraction. The proposed Automatic Extraction and Compensation algorithm is utilized to solve both 16-city traveling salesman problem(TSP) and 21-bit integer factorization on a large P-Bit array with variation, demonstrating its accuracy, transferability, and scalability.

cond-mat.mes-hall↗

arXiv2022

Time Division Multiplexing Ising Computer Using Single Tunable True Random Number Generator Based on Spin Torque Nano-Oscillator

Ising computer is a powerful computation scheme to deal with NP-hard optimization problems that cannot be efficiently addressed by conventional computers. A robust probabilistic bit (P-Bit) which is realized by a hardware entity fluctuating in time between -1 and 1 plays a key role in the success of Ising computer. Spintronics technology, such as stochastic nanomagnet, is recently proposed as a good platform for the hardware emulation of P-Bit. Here, we report, for the first time, a Time Division Multiplexing (TDM) Ising computer using single tunable true random number generator which is comprised of a Spin Torque Nano-Oscillator (STNO). First, the intrinsic frequency fluctuation of the STNO is utilized to design a simple digital true random number generator (TRNG). The true random number generator is further evolved into a tunable random number generator to act as a P-Bit. Second, in order to accomplish combinational optimization with our proposed P-Bit under Ising model, a novel incremental coupling rule is proposed. With such coupling rule between P-Bit array, high fidelity NOT and XOR logic gate is demonstrated. Third, it is proposed that our digital TRNG can be simply reused acting as a P-Bit array by time division multiplexing. The whole Ising computer can be implemented by one single STNO, and integer factorization of as high as 87% accuracy rate is achieved.

cond-mat.mes-hall↗

arXiv2019

Frequency Modulation and Voltage Locking of the Voltage Controlled Spin Oscillators (VCSOs)

The oscillating frequency of typical Spin Torque Nano Oscillators (STNOs) can be modulated by injected DC current or bias magnetic field. And phase locking of STNOs to an external Radio Frequency (RF) signal can be imposed by AC current or RF bias magnetic field. However, in this study, we have proposed a Voltage Controlled Spin Oscillators (VCSOs) by introducing Voltage Controlled Magnetic Anisotropy (VCMA) effect. The oscillating frequency of VCSOs can be modulated by VCMA voltage as well as injected DC current. Furthermore, we have shown a novel locking mechanism caused by AC VCMA voltage. Both the frequency modulation and voltage locking mechanism are analyzed theoretically by Nonlinear Auto-oscillator theory and verified by numerical simulation. At last, we proposed that by utilizing negative capacitance material to enhance VCMA effect, the locking range for voltage locking can be expanded thus may lead to easy mutual synchronization of multiple VCSOs.

cond-mat.mes-hall↗

arXiv2016

Amplicification of Voltage Controlled Magnetic Anisotropy Effect with Negative Capacitance

The high current density required by Magnetic Tunneling Junction (MTJ) switching driven by Spin Transfer Torque (STT) effect leads to large power consumption and severe reliability issues therefore hinder the timetable for STT Magnetic Random Access Memory (STT-MRAM) to mass market. By utilizing Voltage Controlled Magnetic Anisotropy (VCMA) effect, the MTJ can be switched by voltage effect and is postulated to achieve ultra-low power (fJ). However, the VCMA coefficient measured in experiments is far too small for MTJ dimension below 100 nm. Here in this work, a novel approach for the amplification of VCMA effect which borrow ideas from negative capacitance is proposed. The feasibility of the proposal is proved by physical simulation and in-depth analysis.

cond-mat.mes-hall↗