arXiv subjects

Ziheng Huang

Publications and source records attributed to Ziheng Huang.

15 recordsLinked to original sources

arXiv2026

Incremental Consistency Execution for Autonomous Intelligent Systems

Long-horizon autonomous intelligent systems rely on heterogeneous components such as large language models, databases, external APIs, and rule engines, while their external states continuously change during execution. Re-executing the entire workflow after every change introduces substantial redundant computation. This paper proposes an incremental consistency execution method based on task fact contracts, field-level dependency masks, and state perturbation result invariant domains. After an initial verified execution, the system constructs conservative invariant domains for critical inputs and uses them to determine whether downstream results can be safely renewed without re-invoking expensive components. When re-execution is required, only the smallest affected output fields are recomputed, and an equivalence barrier prevents unnecessary downstream propagation. A submission-time version consistency gate further ensures the safety of side-effecting actions. Experiments on industrial fault diagnosis, enterprise analytics, and LLM-based multi-tool assistants show that the proposed method significantly reduces expensive component calls and end-to-end latency while maintaining high consistency and low incorrect-reuse rates.

cs.AI↗

arXiv2026

LIGE-GR: A Smooth Leap from Ranking to Generative Recommendation in the LLM Era

The remarkable success of large language models (LLMs) has provided important inspiration for the next generation of recommender systems. Structurally, recommendation and language generation share a similarity: both aim to produce an ordered sequence that optimizes the user's experience. However, how to precisely absorb the essence of the LLM paradigm into mature industrial recommender systems remains an open problem. There are two challenges. First, it is unclear how to incorporate the LLM paradigm -- sequence-level generation and optimization -- into recommendation. Second, real-world recommender systems are mature systems that have been iteratively customized for years around specific products, business constraints, serving infrastructure, and organizational ownership. Replacing such systems wholesale is often technically risky and organizationally disruptive. In this paper, we propose LIGE-GR, a listwise generation and evaluation recommendation framework that upgrades from a traditional ranking system (itemwise recommendation) toward a generative recommendation paradigm. Instead of rebuilding the entire recommendation stack from scratch, LIGE-GR generalizes the existing pointwise recommendation system into a listwise generation system. This allows mature recommender systems to benefit from listwise optimization while preserving compatibility with existing models, value functions, and serving infrastructure. We validate LIGE-GR in short-video recommendation on Instagram Reels and Facebook Video. On these recommendation surfaces, LIGE-GR improves time spent by 1.14 percent on Instagram Reels and 0.72 percent on Facebook Video, while requiring only modest additional inference resources.

cs.LG↗

arXiv2026

Illuminating Visual Identity in Universal Multimodal Embeddings

Universal Multimodal Embeddings (UMEs) aim to unify various modalities and tasks into a shared representation space. In recent years, this field has witnessed substantial progress driven by the development of Multimodal Large Language Models (MLLMs). However, a crucial capability, visual identity discrimination, remains underexplored in existing UME methods, despite its critical role in a wide range of tasks, including instance retrieval, re-identification, and identity preservation in AI-generated content. To bridge this gap, we propose a unified formulation for visual identity discrimination~(VisID) and introduce $\textbf{MVEB}$ ($\textbf{M}$ultimodal $\textbf{V}$isual Identity $\textbf{E}$mbedding $\textbf{B}$enchmark), a large-scale benchmark curated from both real-world and synthetic datasets to support evaluation and training. Furthermore, we present a simple yet effective learning framework that jointly optimizes general multimodal and visual identity representations through a carefully designed identity-aware sampling mechanism. Extensive experiments demonstrate that our approach successfully endows UMEs with strong identity discrimination capability and maintains competitive general multimodal performance. We believe this work not only illuminates a critical yet neglected capability, but also takes a step toward more holistic universal multimodal embeddings. Code and data are available at \href{https://chrisclear3.github.io/MVEB}{MVEB}.

cs.CV↗

arXiv2026

SlimPer: Make Personalization Model Slim and Smart

Transformer-style architectures are increasingly adopted for industrial recommendation systems, yet they inherit a design premise misaligned with the task: generative models rely on per-token autoregressive prediction, which justifies maintaining large intermediate tensors that scale with sequence length. In contrast, recommendation systems produce a single set of relevance scores for each pair without token-level supervision. Leveraging this observation, we propose SlimPer, which reformulates personalized ranking as iterative refinement of a compact, unified knowledge base. At each layer, the model selectively queries raw multi-modal user-side tokens, computes explicit relevance matching scores, and refines the knowledge base, all in O(N) per-layer cost with a fixed-size intermediate representation. As a result, model depth is decoupled from user history length, enabling deeper relevance understanding without proportional growth in compute or memory; request-only optimization further trims memory by sharing a single copy of user-side tokens across all candidate items. SlimPer unifies sparse, dense, and sequence features within a single backbone and provides inherent interpretability through its attention mechanism. Deployed on Instagram Reels and Feed, SlimPer yields measurable improvements in user engagement while streamlining the overall system and enabling effective modeling of 10k+ fine-grained user history events.

cs.IR↗

arXiv2026

VideoBrain: Learning Adaptive Frame Sampling for Long Video Understanding

Long-form video understanding remains challenging for Vision-Language Models (VLMs) due to the inherent tension between computational constraints and the need to capture information distributed across thousands of frames. Existing approaches either sample frames uniformly (risking information loss) or select keyframes in a single pass (with no recovery from poor choices). We propose VideoBrain, an end-to-end framework that enables VLMs to adaptively acquire visual information through learned sampling policies. Our approach features dual complementary agents: a CLIP-based agent for semantic retrieval across the video and a Uniform agent for dense temporal sampling within intervals. Unlike prior agent-based methods that rely on text-only LLMs orchestrating visual tools, our VLM directly perceives frames and reasons about information sufficiency. To prevent models from invoking agents indiscriminately to maximize rewards, we introduce a behavior-aware reward function coupled with a data classification pipeline that teaches the model when agent invocation is genuinely beneficial. Experiments on four long video benchmarks demonstrate that VideoBrain achieves +3.5% to +9.0% improvement over the baseline while using 30-40\% fewer frames, with strong cross-dataset generalization to short video benchmarks. The code is available at https://github.com/junbo-zou/VideoBrain.

cs.CV↗

arXiv2026

TermSight: Making Service Contracts Approachable

Legal contracts govern much of our society, but their specialized language is difficult for non-experts to read. While AI has enabled simplification of complex language, legal contracts pose unique challenges because of their connection to readers' values, ambiguity, and legally binding nature. Based on a formative study (N=20) using Terms of Service (ToS) as example contracts to study challenges in contract reading, we developed TermSight, an intelligent reading interface to probe the opportunities and challenges of designing augmentations for legal text. TermSight guides readers to relevant clauses with color-coded plain-language snippets of information and contextualizes ambiguous language with definitions and hypothetical scenarios. Importantly, TermSight's features always foreground the original, legally-binding contract text (e.g., linking to associated clauses). Our within-subjects study (N=20) demonstrated the opportunities of TermSight in making ToS significantly easier to read and navigate while revealing the challenges of augmenting service contracts such as ToS.

cs.HC↗

arXiv2026

Living Contracts: Beyond Document-Centric Interaction with Legal Agreements

User interaction with legal contracts has been limited to document reading, which is often complicated by complex, ambiguous legal language. We explore possible futures where contract interfaces go beyond single document interfaces to (1) educate users with legal rights not stated in the contract, (2) transform legal language into alternative representations to aid information tasks before, during, and after signing, and (3) proactively supply contractual information at relevant moments. We refer to these future interfaces collectively as Living Contracts. Using residential leases as a case study, we created three design probes representing different possible Living Contracts. A three-part qualitative study (N=18) revealed participants' barriers to interacting with contracts, including interpreting complex language, uncertainty about legal rights, and the pressure to sign quickly. Participants' feedback on the probes highlighted how Living Contracts have the potential to address these challenges and open new design opportunities for human-contract interactions beyond document reading.

cs.HC↗

arXiv2025

CoMA: Compositional Human Motion Generation with Multi-modal Agents

3D human motion generation has seen substantial advancement in recent years. While state-of-the-art approaches have improved performance significantly, they still struggle with complex and detailed motions unseen in training data, largely due to the scarcity of motion datasets and the prohibitive cost of generating new training examples. To address these challenges, we introduce CoMA, an agent-based solution for complex human motion generation, editing, and comprehension. CoMA leverages multiple collaborative agents powered by large language and vision models, alongside a mask transformer-based motion generator featuring body part-specific encoders and codebooks for fine-grained control. Our framework enables generation of both short and long motion sequences with detailed instructions, text-guided motion editing, and self-correction for improved quality. Evaluations on the HumanML3D dataset demonstrate competitive performance against state-of-the-art methods. Additionally, we create a set of context-rich, compositional, and long text prompts, where user studies show our method significantly outperforms existing approaches.

cs.CV↗

arXiv2023

Memory Sandbox: Transparent and Interactive Memory Management for Conversational Agents

The recent advent of large language models (LLM) has resulted in high-performing conversational agents such as chatGPT. These agents must remember key information from an ongoing conversation to provide responses that are contextually relevant to the user. However, these agents have limited memory and can be distracted by irrelevant parts of the conversation. While many strategies exist to manage conversational memory, users currently lack affordances for viewing and controlling what the agent remembers, resulting in a poor mental model and conversational breakdowns. In this paper, we present Memory Sandbox, an interactive system and design probe that allows users to manage the conversational memory of LLM-powered agents. By treating memories as data objects that can be viewed, manipulated, recorded, summarized, and shared across conversations, Memory Sandbox provides interaction affordances for users to manage how the agent should `see' the conversation.

cs.HC↗

arXiv2023

What can Discriminator do? Towards Box-free Ownership Verification of Generative Adversarial Network

In recent decades, Generative Adversarial Network (GAN) and its variants have achieved unprecedented success in image synthesis. However, well-trained GANs are under the threat of illegal steal or leakage. The prior studies on remote ownership verification assume a black-box setting where the defender can query the suspicious model with specific inputs, which we identify is not enough for generation tasks. To this end, in this paper, we propose a novel IP protection scheme for GANs where ownership verification can be done by checking outputs only, without choosing the inputs (i.e., box-free setting). Specifically, we make use of the unexploited potential of the discriminator to learn a hypersphere that captures the unique distribution learned by the paired generator. Extensive evaluations on two popular GAN tasks and more than 10 GAN architectures demonstrate our proposed scheme to effectively verify the ownership. Our proposed scheme shown to be immune to popular input-based removal attacks and robust against other existing attacks. The source code and models are available at https://github.com/AbstractTeen/gan_ownership_verification

cs.CV↗

arXiv2023

Prompt Middleware: Mapping Prompts for Large Language Models to UI Affordances

To help users do complex work, researchers have developed techniques to integrate AI and human intelligence into user interfaces (UIs). With the recent introduction of large language models (LLMs), which can generate text in response to a natural language prompt, there are new opportunities to consider how to integrate LLMs into UIs. We present Prompt Middleware, a framework for generating prompts for LLMs based on UI affordances. These include prompts that are predefined by experts (static prompts), generated from templates with fill-in options in the UI (template-based prompts), or created from scratch (free-form prompts). We demonstrate this framework with FeedbackBuffet, a writing assistant that automatically generates feedback based on a user's text input. Inspired by prior research showing how templates can help non-experts perform more like experts, FeedbackBuffet leverages template-based prompt middleware to enable feedback seekers to specify the types of feedback they want to receive as options in a UI. These options are composed using a template to form a feedback request prompt to GPT-3. We conclude with a discussion about how Prompt Middleware can help developers integrate LLMs into UIs.

cs.HC↗

arXiv2023

Freeform Templates: Combining Freeform Curation with Structured Templates

Online whiteboards are becoming a popular way to facilitate collaborative design work, providing a free-form environment to curate ideas. However, as templates are increasingly being used to scaffold contributions from non-experts designers, it is crucial to understand their impact on the creative process. In this paper, we present the results from a study with 114 students in a large introductory design course. Our results confirm prior findings that templates benefit students by providing a starting point, a shared process, and the ability to access their own work from previous steps. While prior research has criticized templates for being too rigid, we discovered that using templates within a free-form environment resulted in visual patterns of free-form curation where concepts were spatially organized, clustered, color-coded, and connected using arrows and lines. We introduce the concept of "Free-form Templates" to illustrate how templates and free-form curation can be synergistic.

cs.HC↗

arXiv2022

Heavy-Tailed Loss Frequencies from Mixtures of Negative Binomial and Poisson Counts

Heavy-tailed random variables have been used in insurance research to model both loss frequencies and loss severities, with substantially more emphasis on the latter. In the present work, we take a step toward addressing this imbalance by exploring the class of heavy-tailed frequency models formed by continuous mixtures of Negative Binomial and Poisson random variables. We begin by defining the concept of a calibrative family of mixing distributions (each member of which is identifiable from its associated Negative Binomial mixture), and show how to construct such families from only a single member. We then introduce a new heavy-tailed frequency model -- the two-parameter ZY distribution -- as a generalization of both the one-parameter Zeta and Yule distributions, and construct calibrative families for both the new distribution and the heavy-tailed two-parameter Waring distribution. Finally, we pursue natural extensions of both the ZY and Waring families to a unifying, four-parameter heavy-tailed model, providing the foundation for a novel loss-frequency modeling approach to complement conventional GLM analyses. This approach is illustrated by application to a classic set of Swedish commercial motor-vehicle insurance loss data.

stat.ME↗

arXiv2022

Anti-Forgery: Towards a Stealthy and Robust DeepFake Disruption Attack via Adversarial Perceptual-aware Perturbations

DeepFake is becoming a real risk to society and brings potential threats to both individual privacy and political security due to the DeepFaked multimedia are realistic and convincing. However, the popular DeepFake passive detection is an ex-post forensics countermeasure and failed in blocking the disinformation spreading in advance. To address this limitation, researchers study the proactive defense techniques by adding adversarial noises into the source data to disrupt the DeepFake manipulation. However, the existing studies on proactive DeepFake defense via injecting adversarial noises are not robust, which could be easily bypassed by employing simple image reconstruction revealed in a recent study MagDR. In this paper, we investigate the vulnerability of the existing forgery techniques and propose a novel \emph{anti-forgery} technique that helps users protect the shared facial images from attackers who are capable of applying the popular forgery techniques. Our proposed method generates perceptual-aware perturbations in an incessant manner which is vastly different from the prior studies by adding adversarial noises that is sparse. Experimental results reveal that our perceptual-aware perturbations are robust to diverse image transformations, especially the competitive evasion technique, MagDR via image reconstruction. Our findings potentially open up a new research direction towards thorough understanding and investigation of perceptual-aware adversarial attack for protecting facial images against DeepFakes in a proactive and robust manner. We open-source our tool to foster future research. Code is available at https://github.com/AbstractTeen/AntiForgery/.

cs.CR↗

arXiv2021

Characterizing the Zeta Distribution via Continuous Mixtures

We offer two novel characterizations of the Zeta distribution: first, as tractable continuous mixtures of Negative Binomial distributions (with fixed shape parameter, r > 0), and second, as a tractable continuous mixture of Poisson distributions. In both the Negative Binomial case for r >= 1 and the Poisson case, the resulting Zeta distributions are identifiable because each mixture can be associated with a unique mixing distribution. In the Negative Binomial case for 0 < r < 1, the mixing distributions are quasi-distributions (for which the quasi-probability density function assumes some negative values).

math.PR↗