arXiv subjects

Bing He

Publications and source records attributed to Bing He.

At least 37 records · Page 2Linked to original sources

arXiv2026

How Do Latent Reasoning Methods Perform Under Weak and Strong Supervision?

Latent reasoning has been recently proposed as a reasoning paradigm and performs multi-step reasoning through generating steps in the latent space instead of the textual space. This paradigm enables reasoning beyond discrete language tokens by performing multi-step computation in continuous latent spaces. Although there have been numerous studies focusing on improving the performance of latent reasoning, its internal mechanisms remain not fully investigated. In this work, we conduct a comprehensive analysis of latent reasoning methods to better understand the role and behavior of latent representation in the process. We identify two key issues across latent reasoning methods with different levels of supervision. First, we observe pervasive shortcut behavior, where they achieve high accuracy without relying on latent reasoning. Second, we examine the hypothesis that latent reasoning supports BFS-like exploration in latent space, and find that while latent representations can encode multiple possibilities, the reasoning process does not faithfully implement structured search, but instead exhibits implicit pruning and compression. Finally, our findings reveal a trade-off associated with supervision strength: stronger supervision mitigates shortcut behavior but restricts the ability of latent representations to maintain diverse hypotheses, whereas weaker supervision allows richer latent representations at the cost of increased shortcut behavior.

cs.AI↗

arXiv2026

SurfSplat: Conquering Feedforward 2D Gaussian Splatting with Surface Continuity Priors

Reconstructing 3D scenes from sparse images remains a challenging task due to the difficulty of recovering accurate geometry and texture without optimization. Recent approaches leverage generalizable models to generate 3D scenes using 3D Gaussian Splatting (3DGS) primitive. However, they often fail to produce continuous surfaces and instead yield discrete, color-biased point clouds that appear plausible at normal resolution but reveal severe artifacts under close-up views. To address this issue, we present SurfSplat, a feedforward framework based on 2D Gaussian Splatting (2DGS) primitive, which provides stronger anisotropy and higher geometric precision. By incorporating a surface continuity prior and a forced alpha blending strategy, SurfSplat reconstructs coherent geometry together with faithful textures. Furthermore, we introduce High-Resolution Rendering Consistency (HRRC), a new evaluation metric designed to evaluate high-resolution reconstruction quality. Extensive experiments on RealEstate10K, DL3DV, and ScanNet demonstrate that SurfSplat consistently outperforms prior methods on both standard metrics and HRRC, establishing a robust solution for high-fidelity 3D reconstruction from sparse inputs. Project page: https://hebing-sjtu.github.io/SurfSplat-website/

cs.CV↗

arXiv2025

Proof of a conjecture of Baruah and Sarma on sign patterns of certain infinite products

Let \[ \sum_{n=0}^{\infty}A(n)q^{n} := \frac{(q^{2};q^{5})_{\infty}^{5}(q^{3};q^{5})_{\infty}^{5}}{(q;q^{5})_{\infty}^{5}(q^{4};q^{5})_{\infty}^{5}}, \] \[ \sum_{n=0}^{\infty} B(n)q^{n} := \frac{(q;q^{5})_{\infty}^{5} (q^{4};q^{5})_{\infty}^{5}} {(q^{2};q^{5})_{\infty}^{5}(q^{3}; q^{5})_{\infty}^{5}}, \] and \[ \sum_{n=0}^{\infty} D(n)q^{n} := \frac{(q^{5};q^{25})_{\infty}(q^{20}; q^{25})_{\infty}} {(q^{10};q^{25})_{\infty}(q^{15}; q^{25})_{\infty}} \frac{(q^{2}; q^{5})_{\infty}^{5}(q^{3};q^{5})_{\infty}^{5}} {(q;q^{5})_{\infty}^{5} (q^{4};q^{5})_{\infty}^{5}} \] where $(a;q)_{\infty} := \prod_{k=0}^{\infty}(1-aq^{k})$ and $|q|<1.$ These sequences are closely related to the celebrated Rogers-Ramanujan continued fraction. In this paper, we study the sign behavior o of the coefficients $A(n),B(n)$ and $D(n).$ We prove that for all integers $n\geq0,$ \begin{align*} A(5n)<0\quad(n\neq0),\qquad B(5n) < 0\quad(n\neq0),\qquad D(5n+1)>0. \end{align*} This confirms a recent conjecture of Baruah and Sarma. Our proof is different from the previous method of Baruah and Sarma, and combines asymptotic coefficient analysis with symbolic computation for finite case verification.

math.NT↗

arXiv2025

H3D-DGS: Exploring Heterogeneous 3D Motion Representation for Deformable 3D Gaussian Splatting

Dynamic scene reconstruction poses a persistent challenge in 3D vision. Deformable 3D Gaussian Splatting has emerged as an effective method for this task, offering real-time rendering and high visual fidelity. This approach decomposes a dynamic scene into a static representation in a canonical space and time-varying scene motion. Scene motion is defined as the collective movement of all Gaussian points, and for compactness, existing approaches commonly adopt implicit neural fields or sparse control points. However, these methods predominantly rely on gradient-based optimization for all motion information. Due to the high degree of freedom, they struggle to converge on real-world datasets exhibiting complex motion. To preserve the compactness of motion representation and address convergence challenges, this paper proposes heterogeneous 3D control points, termed \textbf{H3D control points}, whose attributes are obtained using a hybrid strategy combining optical flow back-projection and gradient-based methods. This design decouples directly observable motion components from those that are geometrically occluded. Specifically, components of 3D motion that project onto the image plane are directly acquired via optical flow back projection, while unobservable portions are refined through gradient-based optimization. Experiments on the Neu3DV and CMU-Panoptic datasets demonstrate that our method achieves superior performance over state-of-the-art deformable 3D Gaussian splatting techniques. Remarkably, our method converges within just 100 iterations and achieves a per-frame processing speed of 2 seconds on a single NVIDIA RTX 4070 GPU.

cs.CV↗

arXiv2025

TRAJECT-Bench:A Trajectory-Aware Benchmark for Evaluating Agentic Tool Use

Large language model (LLM)-based agents increasingly rely on tool use to complete real-world tasks. While existing works evaluate the LLMs' tool use capability, they largely focus on the final answers yet overlook the detailed tool usage trajectory, i.e., whether tools are selected, parameterized, and ordered correctly. We introduce TRAJECT-Bench, a trajectory-aware benchmark to comprehensively evaluate LLMs' tool use capability through diverse tasks with fine-grained evaluation metrics. TRAJECT-Bench pairs high-fidelity, executable tools across practical domains with tasks grounded in production-style APIs, and synthesizes trajectories that vary in breadth (parallel calls) and depth (interdependent chains). Besides final accuracy, TRAJECT-Bench also reports trajectory-level diagnostics, including tool selection and argument correctness, and dependency/order satisfaction. Analyses reveal failure modes such as similar tool confusion and parameter-blind selection, and scaling behavior with tool diversity and trajectory length where the bottleneck of transiting from short to mid-length trajectories is revealed, offering actionable guidance for LLMs' tool use.

cs.AI↗

arXiv2025

Adaptive Test-Time Reasoning via Reward-Guided Dual-Phase Search

Large Language Models (LLMs) have achieved significant advances in reasoning tasks. A key approach is tree-based search with verifiers, which expand candidate reasoning paths and use reward models to guide pruning and selection. Although effective in improving accuracy, these methods are not optimal in terms of efficiency: they perform simple decomposition on the reasoning process, but ignore the planning-execution nature of tasks such as math reasoning or code generation. This results in inefficient exploration of reasoning process. To address this, we propose a dual-phase test-time scaling framework that explicitly separates reasoning into planning and execution, and performs search over the two phases individually. Specifically, we decompose reasoning trajectories and develop reward models for each phase, enabling the search to explore and prune plans and executions separately. We further introduce a dynamic budget allocation mechanism that adaptively redistributes sampling effort based on reward feedback, allowing early stopping on confident steps and reallocation of computation to more challenging parts of the reasoning process. Experiments on both mathematical reasoning and code generation benchmarks demonstrate that our approach consistently improves accuracy while reducing redundant computation.

cs.AI↗

arXiv2025

Proof of a conjecture of Garvan and Jennings-Shaffer on the nonnegativity of M_{C1}(m,n) and M_{C5}(m,n)

In their 2016 paper on exotic Bailey--Slater SPT-functions, Garvan and Jennings-Shaffer introduced many new spt-crank-type functions and proposed a conjecture that the spt-crank-type functions $M_{C1}(m,n)$ and $M_{C5}(m,n)$ are both nonnegative for all $m\in\mathbb{Z}$ and $n\in\mathbb{N}.$ Applying Wright\textquoteright s circle method, Jang and Kim showed that $M_{C1}(m,n)$ and $M_{C5}(m,n)$ are both positive for a fixed integer $m$ and large enough integers $n.$ Up to now, no complete proof of this conjecture has been given. In this paper, we provide a complete proof for this conjecture by using the theory of lattice points. Our proof is quite different from that of Jang and Kim.

math.NT↗

arXiv2025

Steering Protein Language Models

Protein Language Models (PLMs), pre-trained on extensive evolutionary data from natural proteins, have emerged as indispensable tools for protein design. While powerful, PLMs often struggle to produce proteins with precisely specified functionalities or properties due to inherent challenges in controlling their outputs. In this work, we investigate the potential of Activation Steering, a technique originally developed for controlling text generation in Large Language Models (LLMs), to direct PLMs toward generating protein sequences with targeted properties. We propose a simple yet effective method that employs activation editing to steer PLM outputs, and extend this approach to protein optimization through a novel editing site identification module. Through comprehensive experiments on lysozyme-like sequence generation and optimization, we demonstrate that our methods can be seamlessly integrated into both auto-encoding and autoregressive PLMs without requiring additional training. These results highlight a promising direction for precise protein engineering using foundation models.

q-bio.BM↗

arXiv2025

Some conjectures of Schlosser and Zhou on sign patterns of the coefficients of infinite products

Recently, Schlosser and Zhou proposed many conjectures on sign patterns of the coefficients appearing in the $q$-series expansions of the infinite Borwein product and other infinite products raised to a real power. In this paper, we will study several of these conjectures. Let \[ G(q):=\prod_{i=1}^{I}\left(\prod_{k=0}^{\infty}(1-q^{m_{i}+kn_{i}})(1-q^{-m_{i}+(k+1)n_{i}})\right)^{u_{i}} \] where $I$ is a positive integer, $1\leq m_{i}<n_{i}$ and $u_{i}\neq0$ for $1\leq i\leq I$ and $|q|<1.$ We will establish an asymptotic formula for the coefficients of $G(q)^δ$ with $δ$ being a positive real number by using the Hardy--Ramanujan--Rademacher circle method. As applications, we apply the asymptotic formula to confirm some of the conjectures of Schlosser and Zhou.

math.CO↗

arXiv2025

Weighted partial sums of a random multiplicative function and their positivity

In this paper, we study the probability that some weighted partial sums of a random multiplicative function $f$ are positive. Applying the characteristic decomposition, we obtain that if $S$ is a non-empty subset of the multiplicative residue class group $(\mathbb{Z}/m\mathbb{Z})^{\times}$ with $m$ being a fixed positive integer and $A=\{a+mn\mid n=0,1,2,3,\cdots\}$ with $a\in S,$ then there exists a positive number $δ$ independent of $x$, such that \[ \mathbb{P}\left(\sum_{A\cap[1,x)}\frac{f(n)}{n}<0\right)>δ\] unless the coefficients of the real characters in the expansion of the characteristic function of $S$ according to the characters of $(\mathbb{Z}/m\mathbb{Z})^{\times}$ are all non-negative, and the coefficients of the complex characters are all zero, in which case we have \[ \mathbb{P}\left(\sum_{A\cap[1,x)}\frac{f(n)}{n}<0\right)=O\left(\exp\left(-\exp\left(\frac{\ln x}{C\ln_{2}x}\right)\right)\right) \] for a positive constant $C.$ This includes as a special case a result of Angelo and Xu. We also extend the result to the cyclotomic field $K_{n}=\mathbb{Q}(ζ_{n})$ with $ζ_{n}=e^{2πi/n}$ and study the probability that these generalized weighted sums are positive. In addition, we deal with the positivity problem of certain partial sums related to the celebrated Ramanujan tau function $τ(n)$ and the Ramanujan modular form $Δ(q),$ and obtain an upper bound for the probability that these partial sums are negative in a more general situation.

math.NT↗

arXiv2025

ROBAD: Robust Adversary-aware Local-Global Attended Bad Actor Detection Sequential Model

Detecting bad actors is critical to ensure the safety and integrity of internet platforms. Several deep learning-based models have been developed to identify such users. These models should not only accurately detect bad actors, but also be robust against adversarial attacks that aim to evade detection. However, past deep learning-based detection models do not meet the robustness requirement because they are sensitive to even minor changes in the input sequence. To address this issue, we focus on (1) improving the model understanding capability and (2) enhancing the model knowledge such that the model can recognize potential input modifications when making predictions. To achieve these goals, we create a novel transformer-based classification model, called ROBAD (RObust adversary-aware local-global attended Bad Actor Detection model), which uses the sequence of user posts to generate user embedding to detect bad actors. Particularly, ROBAD first leverages the transformer encoder block to encode each post bidirectionally, thus building a post embedding to capture the local information at the post level. Next, it adopts the transformer decoder block to model the sequential pattern in the post embeddings by using the attention mechanism, which generates the sequence embedding to obtain the global information at the sequence level. Finally, to enrich the knowledge of the model, embeddings of modified sequences by mimicked attackers are fed into a contrastive-learning-enhanced classification layer for sequence prediction. In essence, by capturing the local and global information (i.e., the post and sequence information) and leveraging the mimicked behaviors of bad actors in training, ROBAD can be robust to adversarial attacks. Extensive experiments on Yelp and Wikipedia datasets show that ROBAD can effectively detect bad actors when under state-of-the-art adversarial attacks.

cs.LG↗

arXiv2025

Memory effects in pulsed optomechanical systems

Memory, understood as time non-locality, is a fundamental property of any physical system, whether classical or quantum, and has important applications in a wide variety of technologies. In the context of quantum technologies, systems with memory can be used in quantum information, communication, and sensing. Here, we demonstrate that cavity optomechanical systems driven by a pulsed laser can operate as programmable quantum memory elements. By engineering the adiabatic and non-adiabatic pulses, particularly the Gaussian and sinusoidal, we induce and control diverse memory phenomena such as dynamical hysteresis, quantized phononic transitions, and distinct energy-storing responses. Within a mean-field approach, we derive the analytical and numerical criteria under which the photonic and phononic observables manifest the memory effects in strongly driven regimes. The memory effects are quantified through a dimensionless geometric form factor, which provides a versatile metric to characterize the memory efficiency. Our protocol is readily compatible with the current optomechanical platforms, highlighting the new possibilities for advanced memory functionalities in quantum technologies.

quant-ph↗

arXiv2025

S2CFormer: Revisiting the RD-Latency Trade-off in Transformer-based Learned Image Compression

Transformer-based Learned Image Compression (LIC) suffers from a suboptimal trade-off between decoding latency and rate-distortion (R-D) performance. Moreover, the critical role of the FeedForward Network (FFN)-based channel aggregation module has been largely overlooked. Our research reveals that efficient channel aggregation-rather than complex and time-consuming spatial operations-is the key to achieving competitive LIC models. Based on this insight, we initiate the ``S2CFormer'' paradigm, a general architecture that simplifies spatial operations and enhances channel operations to overcome the previous trade-off. We present two instances of the S2CFormer: S2C-Conv, and S2C-Attention. Both models demonstrate state-of-the-art (SOTA) R-D performance and significantly faster decoding speed. Furthermore, we introduce S2C-Hybrid, an enhanced variant that maximizes the strengths of different S2CFormer instances to achieve a better performance-latency trade-off. This model outperforms all the existing methods on the Kodak, Tecnick, and CLIC Professional Validation datasets, setting a new benchmark for efficient and high-performance LIC. The code is at \href{https://github.com/YunuoChen/S2CFormer}{https://github.com/YunuoChen/S2CFormer}.

cs.CV↗

arXiv2025

Room temperature mass sensing based on nonlinear optomechanical dynamics: membrane-in-the-middle versus suspended membrane

How to weigh something as precise as possible is a constant endeavor for human being, and mass sensing has been essential to scientific research and many other aspects of modern society. In this work, we explore a special approach to mass sensing, which is purely based on the classical nonlinear dynamics of cavity optomechanical systems. We consider two types of systems, the mechanical resonator as a suspended membrane inside optical cavity or as a larger movable membrane that separates the optical cavity into two parts. Under a driving laser field with two tones satisfying a specific frequency condition, both systems enter a special dynamical pattern correlating the mechanical oscillation and the sidebands of oscillatory cavity field. After adding the nano-particle, which has its mass δm to be measured, to the mechanical membrane as the detector, the cavity field sidebands will exhibit detectable changes, so that the tiny mass δm can be deduced from the measured sideband intensities. For the latter system with a membrane in the middle, one can apply an additional single-tone laser field to magnify the modified sidebands much further, achieving an ultra-high sensitivity (δm/m) \sim 10^{-11} ($m$ is the mass of the membrane), even given a moderate mechanical quality factor. The operation range of the sensors is very wide, covering 7 or 8 orders of magnitudes. Moreover, a particular advantage of this type of mass sensors comes from the robustness of the realized dynamical pattern against thermal noise, and it enables such mass sensors to work well at room temperature.

physics.optics↗

arXiv2024

A Survey on the Role of Crowds in Combating Online Misinformation: Annotators, Evaluators, and Creators

Online misinformation poses a global risk with significant real-world consequences. To combat misinformation, current research relies on professionals like journalists and fact-checkers for annotating and debunking misinformation, and develops automated machine learning methods for detecting misinformation. Complementary to these approaches, recent research has increasingly concentrated on utilizing the power of ordinary social media users, a.k.a. "crowd", who act as eyes-on-the-ground proactively questioning and countering misinformation. Notably, recent studies show that 96% of counter-misinformation responses originate from them. Acknowledging their prominent role, we present the first systematic and comprehensive survey of research papers that actively leverage the crowds to combat misinformation. We first identify 88 papers related to crowd-based efforts, following a meticulous annotation process adhering to the PRISMA framework. We then present key statistics related to misinformation, counter-misinformation, and crowd input in different formats and topics. Upon holistic analysis of the papers, we introduce a novel taxonomy of the roles played by the crowds: (i)annotators who actively identify misinformation; (ii)evaluators who assess counter-misinformation effectiveness; (iii)creators who create counter-misinformation. This taxonomy explores the crowd's capabilities in misinformation detection, identifies prerequisites for effective counter-misinformation, and analyzes crowd-generated counter-misinformation. Then, we delve into (i)distinguishing individual, collaborative, and machine-assisted labeling for annotators; (ii)analyzing the effectiveness of counter-misinformation through surveys, interviews, and in-lab experiments for evaluators; and (iii)characterizing creation patterns and creator profiles for creators. Finally, we outline potential future research in this field.

cs.SI↗

arXiv2024

A scaling law in optomechanically induced nonlinear oscillation

Stable limit cycle as a stabilized mechanical oscillation is the primary result of the dynamical evolution of an optomechanical system under sufficiently powerful pump. Because this dynamical process is highly nonlinear, it was not clear whether there exists a quantitative law to relate an evolved mechanical oscillation (the limit cycle of the dynamical process) to the given parameters of the fabricated system. Here, by means of the numerical simulations based on nonlinear dynamics, we demonstrate the existence of such quantitative relations that are generally valid to the nonlinear optomechanical processes. These quantitative relations can be summarized to a scaling law that is seemingly similar to those in phase transitions of many-body systems but has very different properties. Such a quantitative law enables one to find the more feasible system parameters for realizing the same or a similar dynamical evolution result, so it will be useful to the relevant experimental researches.

physics.optics↗

arXiv2024

Simultaneous photon and phonon lasing in a two-tone driven optomechanical system

Achieving simultaneous lasing of photons and phonons in optomechanical setups has great potential for applications in quantum information processing, high precision sensing and the design of hybrid photonic-phononic devices. Here, we explore this possibility with an optomechanical system driven by a two-tone field. Whenever the difference between the driving frequencies matches the associated mechanical frequency, the photon and phonon populations are found to achieve steady-state coherent oscillations, demonstrating a dual lasing phenomenon. Such drive-tone resonance condition can synchronize the phases of the photon and phonon fields, which facilitates a robust simultaneous lasing. Here, we provide analytical insights into the joint amplification of the optical and mechanical modes, and further confirm the dual lasing phenomenon by numerically calculating the relevant correlation functions and the power spectrum. Our setup, consisting of a single optomechanical cavity, is simpler than previous realizations of dual lasing and provides a clean picture of the underlying mechanisms. Our work thus paves the way for the development of novel strategies for the optimisation of optomechanical interactions through tailored driving schemes.

quant-ph↗

arXiv2024

Learning the irreversible progression trajectory of Alzheimer's disease

Alzheimer's disease (AD) is a progressive and irreversible brain disorder that unfolds over the course of 30 years. Therefore, it is critical to capture the disease progression in an early stage such that intervention can be applied before the onset of symptoms. Machine learning (ML) models have been shown effective in predicting the onset of AD. Yet for subjects with follow-up visits, existing techniques for AD classification only aim for accurate group assignment, where the monotonically increasing risk across follow-up visits is usually ignored. Resulted fluctuating risk scores across visits violate the irreversibility of AD, hampering the trustworthiness of models and also providing little value to understanding the disease progression. To address this issue, we propose a novel regularization approach to predict AD longitudinally. Our technique aims to maintain the expected monotonicity of increasing disease risk during progression while preserving expressiveness. Specifically, we introduce a monotonicity constraint that encourages the model to predict disease risk in a consistent and ordered manner across follow-up visits. We evaluate our method using the longitudinal structural MRI and amyloid-PET imaging data from the Alzheimer's Disease Neuroimaging Initiative (ADNI). Our model outperforms existing techniques in capturing the progressiveness of disease risk, and at the same time preserves prediction accuracy.

cs.LG↗