arXiv subjects

Xiaoyi Gu

Publications and source records attributed to Xiaoyi Gu.

16 recordsLinked to original sources

arXiv2026

CALICO: A Human-Centered, Codebook-Aligned System for Annotation

Large language models are increasingly used to scale codebook-based annotation in scientific research, but existing workflows provide limited support for translating domain experts' codebooks into reliable, revisable, and auditable prompts. Prompts are often treated as fixed instructions and hidden from annotators, making it difficult for non-technical domain experts to diagnose and correct model behavior when outputs violate codebook guidelines. In this paper, we present CALICO, a human-centered, codebook-aligned annotation workflow that treats prompts as editable, versioned, and optimizable artifacts. CALICO integrates codebook parsing, prompt generation, result inspection, prompt versioning, natural language human feedback, and label-supervised prompt optimization through existing optimizers such as GEPA, MIPROv2, and OPRO, together with our reflection-based optimizer, ReflectAgent. Empirically, we evaluate CALICO on domain-specific AI-companion chatbot conversation codebooks. Across evaluated dimensions, CALICO improves mean held-out performance by +13.0 and +7.4 absolute points for two coders, respectively. A coder-specificity analysis further suggests that optimized prompts capture coder-specific interpretations rather than only generic codebook clarification. CALICO runs as a web application that takes users from raw codebook materials to inspectable, exportable labels; the website, codebase, and live demo are released at https://calico-annotation.github.io/ under the Apache 2.0 License.

cs.HC↗

arXiv2026

Multi-Objective Ranking for Live-Streaming: Balancing Fresh and Delayed Signals with Segment-Aware Targeting

One of the most challenging problems entertainment live-streaming services face in recommendation systems is that user behaviors are sparse and delayed, and interaction data exhibits bias for different user segments. Unlike e-commerce applications where user actions follow linear sequences, live-streaming viewers engage in multiple concurrent behaviors of watching, chatting, following, and spending, each occurring with varying delays. We address these challenges through three key contributions: 1) a delayed window approach that extends feedback collection beyond immediate responses, 2) a multi-model architecture that combines fresh and delayed signals, and a segment-aware targeting module that optimizes ranking scores differently across user lifecycle stages, and 3) Multi-gate Mixture-of-Experts (MMoE) integration that jointly models correlated targets while reducing model parameters by 41.9% compared to independent models. Online A/B testing demonstrates significant improvements, including a +0.09% increase in Daily Active Viewers (DAV), generating millions more annual active viewer days, and +0.56% increase in highly engaged viewers' capped Average Revenue Per User (ARPU). Viewer-segment targeting achieved an additional +0.15% DAV improvement for newer and less engaged viewers, while MMoE enhancement added +0.08% overall DAV and +0.27% new follows. The proposed system processes ranking requests with low latency, providing a scalable approach for balancing multiple business objectives across diverse user populations. In addition, we tested the multi-model architecture on the Twitch mobile live feed and achieved a +1.12% increase in positive user-channel interactions (clicks, follows, and likes), demonstrating applicability beyond the primary use case.

cs.IR↗

arXiv2026

Deep Learning for Semen Analysis in Male Infertility: Computer Vision, Multimodal Fusion, and Clinical Translation

Male infertility contributes substantially to the global infertility burden, and sperm analysis remains central to diagnosis, treatment planning, and assisted reproductive technology. Conventional semen evaluation, however, is labor-intensive, operator-dependent, and limited by inter- and intra-observer variability, motivating the development of objective and reproducible computational approaches. This review provides a comprehensive and perspective-oriented synthesis of artificial intelligence-driven sperm analysis, with a focus on computer vision, deep learning, multimodal fusion, robustness, and clinical translation. We first review task-specific methods for sperm detection and counting, tracking-based motility assessment, semantic and instance segmentation, morphology and defect classification, functional assessment, and genetic integrity evaluation. We then summarize public datasets, benchmarks, evaluation metrics, and emerging multimodal strategies that integrate microscopic images, time-lapse videos, CASA-derived parameters, DNA integrity assays, and clinical metadata. Beyond algorithmic performance, we discuss key barriers to real-world deployment, including data scarcity, cross-center domain shift, annotation inconsistency, interpretability, uncertainty calibration, privacy-preserving learning, and workflow integration. Finally, we outline a staged clinical translation roadmap spanning technical standardization, multicenter retrospective validation, silent prospective evaluation, human-in-the-loop clinical testing, ART outcome validation, regulatory approval, and post-market monitoring. By organizing the field from task-specific visual recognition to trustworthy multimodal reproductive intelligence, this review highlights both the progress and the unresolved challenges required to translate AI-driven sperm analysis into clinically meaningful decision support.

cs.CV↗

arXiv2026

DeepTrans Studio: Turning Expert Interventions into Shared Team Knowledge in Agentic Translation Workflows

Professional translation is often a team-based process: translators, reviewers, and project managers must coordinate terminology, legal force, and accountability across documents. Yet many LLM-based translation tools treat human corrections as isolated edits. Expert decisions made in one segment or by one member are rarely captured as reusable knowledge for the rest of the team. We present DeepTrans Studio, a collaborative translation workspace that lets professionals intercept selected nodes in an agentic translation workflow, review evidence, revise AI outputs, and save approved decisions to a shared team memory. During the demo, attendees will role-play translators and reviewers, resolve preset terminology and legal-modal risks, and see how their decisions are propagated to downstream segments and surfaced in a teammate's workspace as reusable precedents. The demo illustrates how human interventions in AI-mediated work can become shared, traceable knowledge rather than one-off corrections.

cs.AI↗

arXiv2023

All-condition pulse detection using a magnetic sensor

A plethora of wearable devices have been developed or commercialized for continuous non-invasive monitoring of physiological signals that are crucial for preventive care and management of chronic conditions. However, most of these devices are either sensitive to skin conditions or its interface with the skin due to the requirement that the external stimuli such as light or electrical excitation must penetrate the skin to detect the pulse. This often results in large motion artefacts and unsuitability for certain skin conditions. Here, we demonstrate a simple fingertip-type device which can detect clear pulse signals under all conditions, including fingers covered by opaque substances such as a plaster or nail polish, or fingers immersed in liquid. The device has a very simple structure, consisting of only a pair of magnets and a magnetic sensor. We show through both experiments and simulations that the detected pulsation signals correspond directly to the magnet vibrations caused by blood circulation, and therefore, in addition to heartrate detection, the proposed device can also be potentially used for blood pressure measurement.

eess.SP↗

arXiv2023

RGI: robust GAN-inversion for mask-free image inpainting and unsupervised pixel-wise anomaly detection

Generative adversarial networks (GANs), trained on a large-scale image dataset, can be a good approximator of the natural image manifold. GAN-inversion, using a pre-trained generator as a deep generative prior, is a promising tool for image restoration under corruptions. However, the performance of GAN-inversion can be limited by a lack of robustness to unknown gross corruptions, i.e., the restored image might easily deviate from the ground truth. In this paper, we propose a Robust GAN-inversion (RGI) method with a provable robustness guarantee to achieve image restoration under unknown \textit{gross} corruptions, where a small fraction of pixels are completely corrupted. Under mild assumptions, we show that the restored image and the identified corrupted region mask converge asymptotically to the ground truth. Moreover, we extend RGI to Relaxed-RGI (R-RGI) for generator fine-tuning to mitigate the gap between the GAN learned manifold and the true image manifold while avoiding trivial overfitting to the corrupted input image, which further improves the image restoration and corrupted region mask identification performance. The proposed RGI/R-RGI method unifies two important applications with state-of-the-art (SOTA) performance: (i) mask-free semantic inpainting, where the corruptions are unknown missing regions, the restored background can be used to restore the missing content; (ii) unsupervised pixel-wise anomaly detection, where the corruptions are unknown anomalous regions, the retrieved mask can be used as the anomalous region's segmentation mask.

cs.CV↗

arXiv2022

Exploiting Instance and Variable Similarity to Improve Learning-Enhanced Branching

In many operational applications, it is necessary to routinely find, within a very limited time window, provably good solutions to challenging mixed-integer linear programming (MILP) problems. An example is the Security-Constrained Unit Commitment (SCUC) problem, solved daily to clear the day-ahead electricity markets. Previous research demonstrated that machine learning (ML) methods can produce high-quality heuristic solutions to combinatorial problems, but proving the optimality of these solutions, even with recently-proposed learning-enhanced branching methods, can still be time-consuming. In this paper, we propose a simple modification to improve the performance of learning-enhanced branching methods based on the key observation that, in such operational applications, instances are significantly similar to each other. Specifically, instances typically share the same size and problem structure, with slight differences only on matrix coefficients, right-hand sides and objective function. In addition, certain groups of variables within a given instance are also typically similar to each other. Therefore, unlike previous works in the literature which predicted all branching scores with a single ML model, we propose training separate ML models per variable or per groups of variables, based on their similarity. We evaluate this enhancement on realistic large-scale SCUC instances and we obtain significantly better gap closures than previous works with the same amount of training data.

math.OC↗

arXiv2022

Solving sparse separable bilinear programs using lifted bilinear cover inequalities

Recently, we proposed a class of inequalities called lifted bilinear cover inequalities, which are second-order cone representable convex inequalities, and are valid for a set described by a separable bilinear constraint together with bounds on variables. In this paper, we study the computational potential of these inequalities for separable bilinear optimization problems. We first prove that the semi-definite programming relaxation provides no benefit over the McCormick relaxation for such problems. We then design a simple randomized separation heuristic for lifted bilinear cover inequalities. In our computational experiments, we separate many rounds of these inequalities starting from McCormick's relaxation of instances where each constraint is a separable bilinear constraint set. We demonstrate that there is a significant improvement in the performance of a state-of-the-art global solver in terms of gap closed, when these inequalities are added at the root node compared to when they are not.

math.OC↗

arXiv2021

Lifting convex inequalities for bipartite bilinear programs

The goal of this paper is to derive new classes of valid convex inequalities for quadratically constrained quadratic programs (QCQPs) through the technique of lifting. Our first main result shows that, for sets described by one bipartite bilinear constraint together with bounds, it is always possible to sequentially lift a seed inequality that is valid for a restriction obtained by fixing variables to their bounds, when the lifting is accomplished using affine functions of the fixed variables. In this setting, sequential lifting involves solving a non-convex nonlinear optimization problem each time a variable is lifted, just as in Mixed Integer Linear Programming. To reduce the computational burden associated with this procedure, we develop a framework based on subadditive approximations of lifting functions that permits sequence-independent lifting of seed inequalities for separable bipartite bilinear sets. In particular, this framework permits the derivation of closed-form valid inequalities. We then study a separable bipartite bilinear set where the coefficients form a minimal cover with respect to the right-hand-side. For this set, we introduce a bilinear cover inequality, which is second-order cone representable. We argue that this bilinear cover inequality is strong by showing that it yields a constant-factor approximation of the convex hull of the original set. We study its lifting function and construct a two-slope subadditive upper bound. Using this subadditive approximation, we lift fixed variable pairs in closed-form, thus deriving a lifted bilinear cover inequality that is valid for general separable bipartite bilinear sets with box constraints.

math.OC↗

arXiv2019

Statistical Analysis of Nearest Neighbor Methods for Anomaly Detection

Nearest-neighbor (NN) procedures are well studied and widely used in both supervised and unsupervised learning problems. In this paper we are concerned with investigating the performance of NN-based methods for anomaly detection. We first show through extensive simulations that NN methods compare favorably to some of the other state-of-the-art algorithms for anomaly detection based on a set of benchmark synthetic datasets. We further consider the performance of NN methods on real datasets, and relate it to the dimensionality of the problem. Next, we analyze the theoretical properties of NN-methods for anomaly detection by studying a more general quantity called distance-to-measure (DTM), originally developed in the literature on robust geometric and topological inference. We provide finite-sample uniform guarantees for the empirical DTM and use them to derive misclassification rates for anomalous observations under various settings. In our analysis we rely on Huber's contamination model and formulate mild geometric regularity assumptions on the underlying distribution of the data.

stat.ML↗

arXiv2019

Exact Augmented Lagrangian Duality for Mixed Integer Quadratic Programming

Mixed integer quadratic programming (MIQP) is the problem of minimizing a convex quadratic function over mixed integer points in a rational polyhedron. This paper focuses on the augmented Lagrangian dual (ALD) for MIQP. ALD augments the usual Lagrangian dual with a weighted nonlinear penalty on the dualized constraints. We first prove that ALD will reach a zero duality gap asymptotically as the weight on the penalty goes to infinity under some mild conditions on the penalty function. We next show that a finite penalty weight is enough for a zero gap when we use any norm as the penalty function. Finally, we prove a polynomially bound on the weight on the penalty term to obtain a zero gap.

math.OC↗

arXiv2017

Global Optimization with Orthogonality Constraints via Stochastic Diffusion on Manifold

Orthogonality constrained optimization is widely used in applications from science and engineering. Due to the nonconvex orthogonality constraints, many numerical algorithms often can hardly achieve the global optimality. We aim at establishing an efficient scheme for finding global minimizers under one or more orthogonality constraints. The main concept is based on noisy gradient flow constructed from stochastic differential equations (SDE) on the Stiefel manifold, the differential geometric characterization of orthogonality constraints. We derive an explicit representation of SDE on the Stiefel manifold endowed with a canonical metric and propose a numerically efficient scheme to simulate this SDE based on Cayley transformation with theoretical convergence guarantee. The convergence to global optimizers is proved under second-order continuity. The effectiveness and efficiency of the proposed algorithms are demonstrated on a variety of problems including homogeneous polynomial optimization, computation of stability number, and 3D structure determination from Common Lines in Cryo-EM.

math.OC↗

arXiv2016

Optimizing quantization for Lasso recovery

This letter is focused on quantized Compressed Sensing, assuming that Lasso is used for signal estimation. Leveraging recent work, we provide a framework to optimize the quantization function and show that the recovered signal converges to the actual signal at a quadratic rate as a function of the quantization level. We show that when the number of observations is high, this method of quantization gives a significantly better recovery rate than standard Lloyd-Max quantization. We support our theoretical analysis with numerical simulations.

cs.IT↗

arXiv2015

Methods for Quantized Compressed Sensing

In this paper, we compare and catalog the performance of various greedy quantized compressed sensing algorithms that reconstruct sparse signals from quantized compressed measurements. We also introduce two new greedy approaches for reconstruction: Quantized Compressed Sampling Matching Pursuit (QCoSaMP) and Adaptive Outlier Pursuit for Quantized Iterative Hard Thresholding (AOP-QIHT). We compare the performance of greedy quantized compressed sensing algorithms for a given bit-depth, sparsity, and noise level.

cs.IT↗

arXiv2015

A note on practical approximate projection schemes in signal space methods

Compressive sensing (CS) is a new technology which allows the acquisition of signals directly in compressed form, using far fewer measurements than traditional theory dictates. Recently, many so-called signal space methods have been developed to extend this body of work to signals sparse in arbitrary dictionaries rather than orthonormal bases. In doing so, CS can be utilized in a much broader array of practical settings. Often, such approaches often rely on the ability to optimally project a signal onto a small number of dictionary atoms. Such optimal, or even approximate, projections have been difficult to derive theoretically. Nonetheless, it has been observed experimentally that conventional CS approaches can be used for such projections, and still provide accurate signal recovery. In this letter, we summarize the empirical evidence and clearly demonstrate for what signal types certain CS methods may be used as approximate projections. In addition, we provide theoretical guarantees for such methods for certain sparse signal structures. Our theoretical results match those observed in experimental studies, and we thus establish both experimentally and theoretically that these CS methods can be used in this context. \end{abstract}

math.NA↗

arXiv2014

Practical approximate projection schemes in greedy signal space methods

Compressive sensing (CS) is a new signal acquisition paradigm which shows that far fewer samples are required to reconstruct sparse signals than previously thought. Although most of the literature focuses on signals sparse in a fixed orthonormal basis, recently the Signal Space CoSaMP (SSCoSaMP) greedy method was developed for the reconstruction of signals compressible in arbitrary redundant dictionaries. The algorithm itself needs access to approximate sparse projection schemes, which have been difficult to obtain and analyze. This paper investigates the use of several different projection schemes and catalogs for what types of signals each scheme can successfully be utilized. In addition, we present novel hybrid projection methods which outperform all other schemes on a wide variety of signal classes.

math.NA↗