arXiv subjects

Jinbin Huang

Publications and source records attributed to Jinbin Huang.

15 recordsLinked to original sources

arXiv2026

Kimi K3: Open Frontier Intelligence

We introduce Kimi K3, a 2.8T parameter Mixture-of-Experts model with 104 billion activated parameters, native vision capabilities, and a 1-million-token context window. Kimi K3 is built on Kimi Delta Attention and Attention Residuals, which improve information flow across sequence length and model depth. Together with Stable LatentMoE, which effectively activates 16 of 896 routed experts per token, and refined training and data recipes, these advances yield an approximately 2.5x improvement in overall scaling efficiency over Kimi K2. Post-training highlights reinforcement learning across general, agentic, and coding domains and multiple reasoning-effort levels, enabling compositional generalization and robust long-horizon execution. At 2.8T scale, Kimi K3 is supported by infrastructure advances in multiple areas: algorithm-system co-design for KDA, perfectly balanced expert-parallel training with efficient memory management, million-token agentic RL with persistent rollout and sandbox states, and deployment innovations. Extensive evaluations show that Kimi K3 achieves frontier-level performance across long-horizon coding, agentic, knowledge, reasoning, and vision tasks. While its overall performance still trails the most powerful proprietary models, namely Claude Fable 5 and GPT-5.6 Sol, Kimi K3 consistently outperforms other open and proprietary models evaluated in our suite. We release the full Kimi K3 model weights to facilitate future research and accelerate the broader deployment and adoption of frontier intelligence.

cs.CL↗

arXiv2026

Reward Hacking as Equilibrium under Finite Evaluation

We prove that under five minimal axioms -- multi-dimensional quality, finite evaluation, effective optimization, resource finiteness, and combinatorial interaction -- any optimized AI agent will systematically under-invest effort in quality dimensions not covered by its evaluation system. This result establishes reward hacking as a structural equilibrium, not a correctable bug, and holds regardless of the specific alignment method (RLHF, DPO, Constitutional AI, or others) or evaluation architecture employed. Our framework instantiates the multi-task principal-agent model of Holmstrom and Milgrom (1991) in the AI alignment setting, but exploits a structural feature unique to AI systems -- the known, differentiable architecture of reward models -- to derive a computable distortion index that predicts both the direction and severity of hacking on each quality dimension prior to deployment. We further prove that the transition from closed reasoning to agentic systems causes evaluation coverage to decline toward zero as tool count grows -- because quality dimensions expand combinatorially while evaluation costs grow at most linearly per tool -- so that hacking severity increases structurally and without bound. Our results unify the explanation of sycophancy, length gaming, and specification gaming under a single theoretical structure and yield an actionable vulnerability assessment procedure. We further conjecture -- with partial formal analysis -- the existence of a capability threshold beyond which agents transition from gaming within the evaluation system (Goodhart regime) to actively degrading the evaluation system itself (Campbell regime), providing the first economic formalization of Bostrom's (2014) "treacherous turn."

cs.AI↗

arXiv2026

Efficient Query Rewrite Rule Discovery via Standardized Enumeration and Learning-to-Rank(extend)

Query rewriting is essential for database performance optimization, but existing automated rule enumeration methods suffer from exponential search spaces, severe redundancy, and poor scalability, especially when handling complex query plans with five or more nodes, where a node represents an operator in the plan tree. We present SLER, a scalable system that enables efficient and effective rewrite rule discovery by combining standardized template enumeration with a learning to rank approach. SLER uses standardized templates, abstractions of query plans with operator structures preserved but data specific details removed, to eliminate structural redundancies and drastically reduce the search space. A learn to rank model guides enumeration by pre filtering the most promising template pairs, enabling scalable rule generation for large node templates. Evaluated on over 11000 real world SQL queries from both open source and commercial workloads, SLER has automatically constructed a rewrite rule repository exceeding 1 million rules - the largest empirically validated rewrite rule library to date. Notably, at the scale of one million rules, SLER supports query plan templates with complexity up to channel level depth. This unprecedented scale opens the door to discovering highly intricate transformations across diverse query patterns. Critically, SLER's template driven design and learned ranking mechanism are inherently extensible, allowing seamless integration of new and complex operators, paving the way for next generation optimizers powered by comprehensive, adaptive rule spaces.

cs.DB↗

arXiv2025

ShapleyPipe: Hierarchical Shapley Search for Data Preparation Pipeline Construction

Automated data preparation pipeline construction is critical for machine learning success, yet existing methods suffer from two fundamental limitations: they treat pipeline construction as black-box optimization without quantifying individual operator contributions, and they struggle with the combinatorial explosion of the search space ($N^M$ configurations for N operators and pipeline length M). We introduce ShapleyPipe, a principled framework that leverages game-theoretic Shapley values to systematically quantify each operator's marginal contribution while maintaining full interpretability. Our key innovation is a hierarchical decomposition that separates category-level structure search from operator-level refinement, reducing the search complexity from exponential to polynomial. To make Shapley computation tractable, we develop: (1) a Multi-Armed Bandit mechanism for intelligent category evaluation with provable convergence guarantees, and (2) Permutation Shapley values to correctly capture position-dependent operator interactions. Extensive evaluation on 18 diverse datasets demonstrates that ShapleyPipe achieves 98.1\% of high-budget baseline performance while using 24\% fewer evaluations, and outperforms the state-of-the-art reinforcement learning method by 3.6\%. Beyond performance gains, ShapleyPipe provides interpretable operator valuations ($\rho$=0.933 correlation with empirical performance) that enable data-driven pipeline analysis and systematic operator library refinement.

cs.DB↗

arXiv2025

SoftPipe: A Soft-Guided Reinforcement Learning Framework for Automated Data Preparation

Data preparation is a foundational yet notoriously challenging component of the machine learning lifecycle, characterized by a vast combinatorial search space. While reinforcement learning (RL) offers a promising direction, state-of-the-art methods suffer from a critical limitation: to manage the search space, they rely on rigid ``hard constraints'' that prematurely prune the search space and often preclude optimal solutions. To address this, we introduce SoftPipe, a novel RL framework that replaces these constraints with a flexible ``soft guidance'' paradigm. SoftPipe formulates action selection as a Bayesian inference problem. A high-level strategic prior, generated by a Large Language Model (LLM), probabilistically guides exploration. This prior is combined with empirical estimators from two sources through a collaborative process: a fine-grained quality score from a supervised Learning-to-Rank (LTR) model and a long-term value estimate from the agent's Q-function. Through extensive experiments on 18 diverse datasets, we demonstrate that SoftPipe achieves up to a 13.9\% improvement in pipeline quality and 2.8$\times$ faster convergence compared to existing methods.

cs.DB↗

arXiv2025

LLaPipe: LLM-Guided Reinforcement Learning for Automated Data Preparation Pipeline Construction

Automated data preparation is crucial for democratizing machine learning, yet existing reinforcement learning (RL) based approaches suffer from inefficient exploration in the vast space of possible preprocessing pipelines. We present LLaPipe, a novel framework that addresses this exploration bottleneck by integrating Large Language Models (LLMs) as intelligent policy advisors. Unlike traditional methods that rely solely on statistical features and blind trial-and-error, LLaPipe leverages the semantic understanding capabilities of LLMs to provide contextually relevant exploration guidance. Our framework introduces three key innovations: (1) an LLM Policy Advisor that analyzes dataset semantics and pipeline history to suggest promising preprocessing operations, (2) an Experience Distillation mechanism that mines successful patterns from past pipelines and transfers this knowledge to guide future exploration, and (3) an Adaptive Advisor Triggering strategy (Advisor\textsuperscript{+}) that dynamically determines when LLM intervention is most beneficial, balancing exploration effectiveness with computational cost. Through extensive experiments on 18 diverse datasets spanning multiple domains, we demonstrate that LLaPipe achieves up to 22.4\% improvement in pipeline quality and 2.3$\times$ faster convergence compared to state-of-the-art RL-based methods, while maintaining computational efficiency through selective LLM usage (averaging only 19.0\% of total exploration steps).

cs.DB↗

arXiv2024

InterVLS: Interactive Model Understanding and Improvement with Vision-Language Surrogates

Deep learning models are widely used in critical applications, highlighting the need for pre-deployment model understanding and improvement. Visual concept-based methods, while increasingly used for this purpose, face challenges: (1) most concepts lack interpretability, (2) existing methods require model knowledge, often unavailable at run time. Additionally, (3) there lacks a no-code method for post-understanding model improvement. Addressing these, we present InterVLS. The system facilitates model understanding by discovering text-aligned concepts, measuring their influence with model-agnostic linear surrogates. Employing visual analytics, InterVLS offers concept-based explanations and performance insights. It enables users to adjust concept influences to update a model, facilitating no-code model improvement. We evaluate InterVLS in a user study, illustrating its functionality with two scenarios. Results indicates that InterVLS is effective to help users identify influential concepts to a model, gain insights and adjust concept influence to improve the model. We conclude with a discussion based on our study results.

cs.AI↗

arXiv2024

InFiConD: Interactive No-code Fine-tuning with Concept-based Knowledge Distillation

The emergence of large-scale pre-trained models has heightened their application in various downstream tasks, yet deployment is a challenge in environments with limited computational resources. Knowledge distillation has emerged as a solution in such scenarios, whereby knowledge from large teacher models is transferred into smaller student' models, but this is a non-trivial process that traditionally requires technical expertise in AI/ML. To address these challenges, this paper presents InFiConD, a novel framework that leverages visual concepts to implement the knowledge distillation process and enable subsequent no-code fine-tuning of student models. We develop a novel knowledge distillation pipeline based on extracting text-aligned visual concepts from a concept corpus using multimodal models, and construct highly interpretable linear student models based on visual concepts that mimic a teacher model in a response-based manner. InFiConD's interface allows users to interactively fine-tune the student model by manipulating concept influences directly in the user interface. We validate InFiConD via a robust usage scenario and user study. Our findings indicate that InFiConD's human-in-the-loop and visualization-driven approach enables users to effectively create and analyze student models, understand how knowledge is transferred, and efficiently perform fine-tuning operations. We discuss how this work highlights the potential of interactive and visual methods in making knowledge distillation and subsequent no-code fine-tuning more accessible and adaptable to a wider range of users with domain-specific demands.

cs.LG↗

arXiv2024

ASAP: Interpretable Analysis and Summarization of AI-generated Image Patterns at Scale

Generative image models have emerged as a promising technology to produce realistic images. Despite potential benefits, concerns grow about its misuse, particularly in generating deceptive images that could raise significant ethical, legal, and societal issues. Consequently, there is growing demand to empower users to effectively discern and comprehend patterns of AI-generated images. To this end, we developed ASAP, an interactive visualization system that automatically extracts distinct patterns of AI-generated images and allows users to interactively explore them via various views. To uncover fake patterns, ASAP introduces a novel image encoder, adapted from CLIP, which transforms images into compact "distilled" representations, enriched with information for differentiating authentic and fake images. These representations generate gradients that propagate back to the attention maps of CLIP's transformer block. This process quantifies the relative importance of each pixel to image authenticity or fakeness, exposing key deceptive patterns. ASAP enables the at scale interactive analysis of these patterns through multiple, coordinated visualizations. This includes a representation overview with innovative cell glyphs to aid in the exploration and qualitative evaluation of fake patterns across a vast array of images, as well as a pattern view that displays authenticity-indicating patterns in images and quantifies their impact. ASAP supports the analysis of cutting-edge generative models with the latest architectures, including GAN-based models like proGAN and diffusion models like the latent diffusion model. We demonstrate ASAP's usefulness through two usage scenarios using multiple fake image detection benchmark datasets, revealing its ability to identify and understand hidden patterns in AI-generated images, especially in detecting fake human faces produced by diffusion-based techniques.

cs.CV↗

arXiv2023

PromptAid: Prompt Exploration, Perturbation, Testing and Iteration using Visual Analytics for Large Language Models

Large Language Models (LLMs) have gained widespread popularity due to their ability to perform ad-hoc Natural Language Processing (NLP) tasks with a simple natural language prompt. Part of the appeal for LLMs is their approachability to the general public, including individuals with no prior technical experience in NLP techniques. However, natural language prompts can vary significantly in terms of their linguistic structure, context, and other semantics. Modifying one or more of these aspects can result in significant differences in task performance. Non-expert users may find it challenging to identify the changes needed to improve a prompt, especially when they lack domain-specific knowledge and lack appropriate feedback. To address this challenge, we present PromptAid, a visual analytics system designed to interactively create, refine, and test prompts through exploration, perturbation, testing, and iteration. PromptAid uses multiple, coordinated visualizations which allow users to improve prompts by using the three strategies: keyword perturbations, paraphrasing perturbations, and obtaining the best set of in-context few-shot examples. PromptAid was designed through an iterative prototyping process involving NLP experts and was evaluated through quantitative and qualitative assessments for LLMs. Our findings indicate that PromptAid helps users to iterate over prompt template alterations with less cognitive overhead, generate diverse prompts with help of recommendations, and analyze the performance of the generated prompts while surpassing existing state-of-the-art prompting interfaces in performance.

cs.HC↗

arXiv2022

ConceptExplainer: Interactive Explanation for Deep Neural Networks from a Concept Perspective

Traditional deep learning interpretability methods which are suitable for model users cannot explain network behaviors at the global level and are inflexible at providing fine-grained explanations. As a solution, concept-based explanations are gaining attention due to their human intuitiveness and their flexibility to describe both global and local model behaviors. Concepts are groups of similarly meaningful pixels that express a notion, embedded within the network's latent space and have commonly been hand-generated, but have recently been discovered by automated approaches. Unfortunately, the magnitude and diversity of discovered concepts makes it difficult to navigate and make sense of the concept space. Visual analytics can serve a valuable role in bridging these gaps by enabling structured navigation and exploration of the concept space to provide concept-based insights of model behavior to users. To this end, we design, develop, and validate ConceptExplainer, a visual analytics system that enables people to interactively probe and explore the concept space to explain model behavior at the instance/class/global level. The system was developed via iterative prototyping to address a number of design challenges that model users face in interpreting the behavior of deep learning models. Via a rigorous user study, we validate how ConceptExplainer supports these challenges. Likewise, we conduct a series of usage scenarios to demonstrate how the system supports the interactive analysis of model behavior across a variety of tasks and explanation granularities, such as identifying concepts that are important to classification, identifying bias in training data, and understanding how concepts can be shared across diverse and seemingly dissimilar classes.

cs.HC↗

arXiv2021

Why? Why not? When? Visual Explanations of Agent Behavior in Reinforcement Learning

Reinforcement learning (RL) is used in many domains, including autonomous driving, robotics, stock trading, and video games. Unfortunately, the black box nature of RL agents, combined with legal and ethical considerations, makes it increasingly important that humans (including those are who not experts in RL) understand the reasoning behind the actions taken by an RL agent, particularly in safety-critical domains. To help address this challenge, we introduce PolicyExplainer, a visual analytics interface which lets the user directly query an autonomous agent. PolicyExplainer visualizes the states, policy, and expected future rewards for an agent, and supports asking and answering questions such as: Why take this action? Why not take this other action? When is this action taken? PolicyExplainer is designed based upon a domain analysis with RL researchers, and is evaluated via qualitative and quantitative assessments on a trio of domains: taxi navigation, a stack bot domain, and drug recommendation for HIV patients. We find that PolicyExplainer promotes trust and understanding of agent decisions better than a state-of-the-art text-based explanation approach. Interviews with domain practitioners provide further validation for PolicyExplainer as applied to safety-critical domains. Our results help demonstrate how visualization-based approaches can be leveraged to decode the behavior of autonomous RL agents, particularly for RL non-experts.

cs.HC↗

arXiv2021

BU-Trace: A Permissionless Mobile System for Privacy-Preserving Intelligent Contact Tracing

The coronavirus disease 2019 (COVID-19) pandemic has caused an unprecedented health crisis for the global. Digital contact tracing, as a transmission intervention measure, has shown its effectiveness on pandemic control. Despite intensive research on digital contact tracing, existing solutions can hardly meet users' requirements on privacy and convenience. In this paper, we propose BU-Trace, a novel permissionless mobile system for privacy-preserving intelligent contact tracing based on QR code and NFC technologies. First, a user study is conducted to investigate and quantify the user acceptance of a mobile contact tracing system. Second, a decentralized system is proposed to enable contact tracing while protecting user privacy. Third, an intelligent behavior detection algorithm is designed to ease the use of our system. We implement BU-Trace and conduct extensive experiments in several real-world scenarios. The experimental results show that BU-Trace achieves a privacy-preserving and intelligent mobile system for contact tracing without requesting location or other privacy-related permissions.

cs.SI↗

arXiv2020

Truss-based Structural Diversity Search in Large Graphs

Social decisions made by individuals are easily influenced by information from their social neighborhoods. A key predictor of social contagion is the multiplicity of social contexts inside the individual's contact neighborhood, which is termed structural diversity. However, the existing models have limited decomposability for analyzing large-scale networks, and suffer from the inaccurate reflection of social context diversity. In this paper, we propose a new truss-based structural diversity model to overcome the weak decomposability. Based on this model, we study a novel problem of truss-based structural diversity search in a graph G, that is, to find the r vertices with the highest truss-based structural diversity and return their social contexts. o tackle this problem, we propose an online structural diversity search algorithm in $O(ρ(m+\mathcal{T}))$ time, where $ρ$, $m$, and $\mathcal{T}$ are respectively the arboricity, the number of edges, and the number of triangles in $G$. To improve the efficiency, we design an elegant and compact index, called TSD-index, for further expediting the search process. We further optimize the structure of TSD-index into a highly compressed GCT-index. Our GCT-index-based structural diversity search utilizes the global triangle information for fast index construction and finds answers in $O(m)$ time. Extensive experiments demonstrate the effectiveness and efficiency of our proposed model and algorithms, against state-of-the-art methods.

cs.DB↗

arXiv2019

Parameter-free Structural Diversity Search

The problem of structural diversity search is to find the top-k vertices with the largest structural diversity in a graph. However, when identifying distinct social contexts, existing structural diversity models (e.g., t-sized component, t-core, and t-brace) are sensitive to an input parameter of t. To address this drawback, we propose a parameter-free structural diversity model. Specifically, we propose a novel notation of discriminative core, which automatically models various kinds of social contexts without parameter t. Leveraging on discriminative cores and h-index, the structural diversity score for a vertex is calculated. We study the problem of parameter-free structural diversity search in this paper. An efficient top-k search algorithm with a well-designed upper bound for pruning is proposed. Extensive experiment results demonstrate the parameter sensitivity of existing t-core based model and verify the superiority of our methods.

cs.DB↗