CVPR 2026 Call for Papers
This article appears to be unrelated to Immigration Law practice area. However, I can analyze it for relevance in other areas. Key legal developments, research findings, and policy signals in this article are not applicable to Immigration Law. The article is a call for papers for the Computer Vision and Pattern Recognition (CVPR) 2026 conference, focusing on computer vision and pattern recognition topics. If I were to stretch for relevance, I could say that the article touches on the intersection of technology and society, which might be of interest to immigration lawyers who deal with issues related to technology and global migration, such as the use of biometric data in border control or the impact of artificial intelligence on immigration decision-making. However, this is a very indirect and tenuous connection.
The CVPR 2026 Call for Papers, while focused on computer vision research, indirectly informs immigration law practice by influencing the development of technologies relevant to biometrics, surveillance, and data privacy—areas intersecting with immigration enforcement and border security. In the U.S., advancements in biometric identification may impact regulatory frameworks governing data collection, aligning with evolving privacy laws like the CPRA. South Korea’s stringent biometric data protections under the Personal Information Protection Act similarly shape compliance strategies for immigration-related tech. Internationally, the EU’s GDPR-driven approach underscores a global trend toward balancing innovation with individual rights, creating a shared imperative for legal practitioners to adapt to technological shifts affecting immigration law. Thus, even indirect research forums like CVPR contribute to shaping legal adaptation in immigration contexts.
As the Work Visa & Employment-Based Immigration Expert, I'll provide domain-specific expert analysis of the article's implications for practitioners in the context of H-1B, L-1, O-1, and employment-based green cards. The article's focus on computer vision and pattern recognition is closely related to the O-1 visa category, which requires evidence of extraordinary ability in the field. The list of topics of interest in computer vision and pattern recognition, such as deep learning architectures and techniques, image and video synthesis and generation, and multimodal learning, are areas where O-1 visa applicants may demonstrate their expertise and qualify for the visa. Practitioners should note that the O-1 visa category requires a labor certification or a job offer from a U.S. employer, and the beneficiary must demonstrate extraordinary ability in their field through evidence such as publications, awards, and recognition. The article's emphasis on original research and high-quality papers may be relevant to the O-1 visa process, as applicants may use their research and publications as evidence of their expertise. In terms of case law, statutory, or regulatory connections, the O-1 visa category is governed by 8 CFR 214.2(o), which outlines the requirements for the visa. The statute governing the O-1 visa is 8 U.S.C. 1101(a)(15)(O), which defines the category. The regulatory framework for the O-1 visa is also informed by the Immigration and Nationality Act
AD-Bench: A Real-World, Trajectory-Aware Advertising Analytics Benchmark for LLM Agents
arXiv:2602.14257v1 Announce Type: new Abstract: While Large Language Model (LLM) agents have achieved remarkable progress in complex reasoning tasks, evaluating their performance in real-world environments has become a critical problem. Current benchmarks, however, are largely restricted to idealized simulations, failing...
The article *AD-Bench: A Real-World, Trajectory-Aware Advertising Analytics Benchmark for LLM Agents* is not directly relevant to **Immigration Law practice**, as it focuses on evaluating AI agents in advertising and marketing analytics rather than legal or policy frameworks. However, it may indirectly signal trends in **AI-driven legal tech** and **automated document analysis**, which could eventually intersect with immigration case management or regulatory compliance tools. For now, immigration practitioners should monitor AI advancements in adjacent fields but note that this study does not introduce legal or policy changes affecting immigration practice.
### **Jurisdictional Comparison & Analytical Commentary on AD-Bench’s Impact on Immigration Law Practice** The emergence of **AD-Bench**—a real-world benchmark for evaluating LLM agents in advertising analytics—has broader implications for **immigration law practice**, particularly in **automated legal decision-making, client intake, and document analysis**. While the U.S. and South Korea have taken divergent approaches to AI adoption in legal services, international frameworks (e.g., EU AI Act) provide a comparative lens. 1. **United States Approach**: The U.S. has adopted a **fragmented, case-by-case regulatory approach**, with agencies like USCIS and EOIR gradually integrating AI tools for visa processing and asylum adjudication. AD-Bench’s emphasis on **multi-tool collaboration** (L3 difficulty) mirrors U.S. immigration systems that require cross-referencing multiple databases (e.g., FBI, DHS, Interpol). However, unlike AD-Bench’s structured benchmarking, U.S. immigration AI adoption remains **ad hoc**, with concerns over **bias in algorithmic decision-making** (e.g., *Matter of A-B-*, 2018) leading to calls for stricter oversight. 2. **South Korean Approach**: South Korea’s immigration system, under the **Ministry of Justice (MOJ)**, has been more **centralized in AI adoption**, particularly in biometric screening and visa fraud detection.
As the Work Visa & Employment-Based Immigration Expert, I will analyze the article's implications for practitioners in the context of immigration law. The article discusses the development of a benchmark for evaluating Large Language Model (LLM) agents in complex advertising and marketing analytics tasks. While this may seem unrelated to immigration law, it has implications for practitioners who work with highly skilled foreign workers in specialized fields like data science and artificial intelligence. In the context of H-1B and L-1 visas, which are often used by employers to sponsor foreign workers in specialized fields, the AD-Bench benchmark may be relevant in demonstrating the qualifications and expertise of foreign workers. For instance, an employer may use AD-Bench as a tool to evaluate the abilities of a foreign worker in a data science or AI role, and demonstrate their qualifications to U.S. Citizenship and Immigration Services (USCIS) as part of an H-1B or L-1 petition. In terms of statutory or regulatory connections, this article may be relevant to the discussion around the "specialty occupation" definition in 8 CFR 214.2(h)(4)(ii), which defines a specialty occupation as one that requires theoretical and practical application of a body of highly specialized knowledge. The AD-Bench benchmark may be used to demonstrate that a foreign worker has the necessary expertise and qualifications to work in a specialty occupation, and therefore may be eligible for an H-1B or L-1 visa. Additionally, the article's discussion
Benchmark Leakage Trap: Can We Trust LLM-based Recommendation?
arXiv:2602.13626v1 Announce Type: new Abstract: The expanding integration of Large Language Models (LLMs) into recommender systems poses critical challenges to evaluation reliability. This paper identifies and investigates a previously overlooked issue: benchmark data leakage in LLM-based recommendation. This phenomenon occurs...
The academic article on LLM-based recommendation systems has indirect relevance to Immigration Law practice by highlighting systemic issues in evaluating algorithmic performance—specifically, data leakage in AI models can produce misleading metrics that affect decision-making. Key legal developments include the recognition that algorithmic bias or inaccuracy stemming from hidden data exposure may have implications for regulatory compliance, particularly in areas where AI is used for immigration eligibility assessments or recommendation platforms. The findings signal a growing need for transparency and validation protocols in AI-driven systems, prompting practitioners to consider potential legal risks associated with reliance on AI recommendations in client advising or administrative decision-making.
Jurisdictional Comparison and Analytical Commentary: The phenomenon of benchmark data leakage in Large Language Models (LLMs) poses significant implications for Immigration Law practice, particularly in the realm of asylum and refugee claims. In the US, for instance, the use of AI-powered tools to evaluate asylum claims could be compromised by data leakage, potentially leading to inaccurate determinations of refugee status. In contrast, Korea's approach to AI adoption in immigration decision-making is still in its nascent stages, and it remains to be seen how the country will address the issue of data leakage. Internationally, the International Organization for Migration (IOM) and other humanitarian organizations may need to reassess their reliance on AI-powered tools in refugee resettlement and protection efforts. In the context of Immigration Law, the impact of data leakage on LLM-based recommendation systems could be far-reaching. If LLMs are exposed to and memorize benchmark datasets, it could lead to artificially inflated performance metrics that fail to reflect true model performance. This could result in inaccurate determinations of refugee status, leading to potential human rights violations. Furthermore, the use of AI-powered tools in immigration decision-making raises concerns about transparency, accountability, and the potential for bias. In the US, the use of AI-powered tools in immigration decision-making is governed by the Administrative Procedure Act (APA) and the Immigration and Nationality Act (INA). However, the APA does not specifically address the issue of data leakage, and the INA does not provide clear guidelines for
The article on LLM-based recommendation data leakage raises critical implications for practitioners by highlighting a previously unrecognized vulnerability in evaluating AI performance. Specifically, the issue of data leakage—where LLMs are exposed to benchmark datasets during pre-training or fine-tuning—creates a misleading inflation of performance metrics, potentially distorting the perceived efficacy of models. From an immigration and legal perspective, practitioners advising on AI-related immigration petitions (e.g., O-1 for extraordinary ability or H-1B for specialty occupations) should be cognizant of the potential for inflated claims of AI capabilities due to such leakage, as it may affect the substantiation of expertise or technological innovation in petitions. Statutorily, this aligns with concerns under 8 U.S.C. § 1153(b)(2) regarding the requirement for genuine expertise, and case law such as Matter of Chawla may inform scrutiny of claims tied to AI performance metrics. Practitioners should integrate awareness of these evaluation pitfalls into due diligence for clients.
A Parameter-Efficient Transfer Learning Approach through Multitask Prompt Distillation and Decomposition for Clinical NLP
arXiv:2604.06650v1 Announce Type: new Abstract: Existing prompt-based fine-tuning methods typically learn task-specific prompts independently, imposing significant computing and storage overhead at scale when deploying multiple clinical natural language processing (NLP) systems. We present a multitask prompt distillation and decomposition framework...
AgentOpt v0.1 Technical Report: Client-Side Optimization for LLM-Based Agent
arXiv:2604.06296v1 Announce Type: new Abstract: AI agents are increasingly deployed in real-world applications, including systems such as Manus, OpenClaw, and coding agents. Existing research has primarily focused on \emph{server-side} efficiency, proposing methods such as caching, speculative execution, traffic scheduling, and...
State-of-the-Art Arabic Language Modeling with Sparse MoE Fine-Tuning and Chain-of-Thought Distillation
arXiv:2604.06421v1 Announce Type: new Abstract: This paper introduces Arabic-DeepSeek-R1, an application-driven open-source Arabic LLM that leverages a sparse MoE backbone to address the digital equity gap for under-represented languages, and establishes a new SOTA across the entire Open Arabic LLM...
ART: Attention Replacement Technique to Improve Factuality in LLMs
arXiv:2604.06393v1 Announce Type: new Abstract: Hallucination in large language models (LLMs) continues to be a significant issue, particularly in tasks like question answering, where models often generate plausible yet incorrect or irrelevant information. Although various methods have been proposed to...
DiffuMask: Diffusion Language Model for Token-level Prompt Pruning
arXiv:2604.06627v1 Announce Type: new Abstract: In-Context Learning and Chain-of-Thought prompting improve reasoning in large language models (LLMs). These typically come at the cost of longer, more expensive prompts that may contain redundant information. Prompt compression based on pruning offers a...
Multi-objective Evolutionary Merging Enables Efficient Reasoning Models
arXiv:2604.06465v1 Announce Type: new Abstract: Reasoning models have demonstrated remarkable capabilities in solving complex problems by leveraging long chains of thought. However, this more deliberate reasoning comes with substantial computational overhead at inference time. The Long-to-Short (L2S) reasoning problem seeks...
Distributed Interpretability and Control for Large Language Models
arXiv:2604.06483v1 Announce Type: new Abstract: Large language models that require multiple GPU cards to host are usually the most capable models. It is necessary to understand and steer these models, but the current technologies do not support the interpretability and...
A Severity-Based Curriculum Learning Strategy for Arabic Medical Text Generation
arXiv:2604.06365v1 Announce Type: new Abstract: Arabic medical text generation is increasingly needed to help users interpret symptoms and access general health guidance in their native language. Nevertheless, many existing methods assume uniform importance across training samples, overlooking differences in clinical...
Does a Global Perspective Help Prune Sparse MoEs Elegantly?
arXiv:2604.06542v1 Announce Type: new Abstract: Empirical scaling laws for language models have encouraged the development of ever-larger LLMs, despite their growing computational and memory costs. Sparse Mixture-of-Experts (MoEs) offer a promising alternative by activating only a subset of experts per...
In-Context Learning in Speech Language Models: Analyzing the Role of Acoustic Features, Linguistic Structure, and Induction Heads
arXiv:2604.06356v1 Announce Type: new Abstract: In-Context Learning (ICL) has been extensively studied in text-only Language Models, but remains largely unexplored in the speech domain. Here, we investigate how linguistic and acoustic features affect ICL in Speech Language Models. We focus...
Databricks co-founder wins prestigious ACM award, says ‘AGI is here already’
Matei Zaharia has won the top honor from the Association for Computing Machinery. Now he's working on AI for research and says AGI is simply misunderstood.
Spectral Edge Dynamics Reveal Functional Modes of Learning
arXiv:2604.06256v1 Announce Type: new Abstract: Training dynamics during grokking concentrate along a small number of dominant update directions -- the spectral edge -- which reliably distinguishes grokking from non-grokking regimes. We show that standard mechanistic interpretability tools (head attribution, activation...
The Illusion of Stochasticity in LLMs
arXiv:2604.06543v1 Announce Type: new Abstract: In this work, we demonstrate that reliable stochastic sampling is a fundamental yet unfulfilled requirement for Large Language Models (LLMs) operating as agents. Agentic systems are frequently required to sample from distributions, often inferred from...
The Illusion of Superposition? A Principled Analysis of Latent Thinking in Language Models
arXiv:2604.06374v1 Announce Type: new Abstract: Latent reasoning via continuous chain-of-thoughts (Latent CoT) has emerged as a promising alternative to discrete CoT reasoning. Operating in continuous space increases expressivity and has been hypothesized to enable superposition: the ability to maintain multiple...
SHAPE: Stage-aware Hierarchical Advantage via Potential Estimation for LLM Reasoning
arXiv:2604.06636v1 Announce Type: new Abstract: Process supervision has emerged as a promising approach for enhancing LLM reasoning, yet existing methods fail to distinguish meaningful progress from mere verbosity, leading to limited reasoning capabilities and unresolved token inefficiency. To address this,...
Graph-Based Chain-of-Thought Pruning for Reducing Redundant Reflections in Reasoning LLMs
arXiv:2604.05643v1 Announce Type: new Abstract: Extending CoT through RL has been widely used to enhance the reasoning capabilities of LLMs. However, due to the sparsity of reward signals, it can also induce undesirable thinking patterns such as overthinking, i.e., generating...
IntentScore: Intent-Conditioned Action Evaluation for Computer-Use Agents
arXiv:2604.05157v1 Announce Type: new Abstract: Computer-Use Agents (CUAs) leverage large language models to execute GUI operations on desktop environments, yet they generate actions without evaluating action quality, leading to irreversible errors that cascade through subsequent steps. We propose IntentScore, a...
Top-K Retrieval with Fixed-Size Linear-Attention Completion: Backbone- and KV-Format-Preserving Attention for KV-Cache Read Reduction
arXiv:2604.05438v1 Announce Type: new Abstract: Long-context generation is increasingly limited by decode-time key-value (KV) cache traffic, particularly when KV is offloaded beyond GPU memory. Query-aware retrieval (e.g., Top-K selection) reduces this traffic by loading only a subset of KV pairs,...
Attention Editing: A Versatile Framework for Cross-Architecture Attention Conversion
arXiv:2604.05688v1 Announce Type: new Abstract: Key-Value (KV) cache memory and bandwidth increasingly dominate large language model inference cost in long-context and long-generation regimes. Architectures such as multi-head latent attention (MLA) and hybrid sliding-window attention (SWA) can alleviate this bound, but...
Jeffreys Flow: Robust Boltzmann Generators for Rare Event Sampling via Parallel Tempering Distillation
arXiv:2604.05303v1 Announce Type: new Abstract: Sampling physical systems with rough energy landscapes is hindered by rare events and metastable trapping. While Boltzmann generators already offer a solution, their reliance on the reverse Kullback--Leibler divergence frequently induces catastrophic mode collapse, missing...
ALTO: Adaptive LoRA Tuning and Orchestration for Heterogeneous LoRA Training Workloads
arXiv:2604.05426v1 Announce Type: new Abstract: Low-Rank Adaptation (LoRA) is now the dominant method for parameter-efficient fine-tuning of large language models, but achieving a high-quality adapter often requires systematic hyperparameter tuning because LoRA performance is highly sensitive to configuration choices. In...
XMark: Reliable Multi-Bit Watermarking for LLM-Generated Texts
arXiv:2604.05242v1 Announce Type: new Abstract: Multi-bit watermarking has emerged as a promising solution for embedding imperceptible binary messages into Large Language Model (LLM)-generated text, enabling reliable attribution and tracing of malicious usage of LLMs. Despite recent progress, existing methods still...
CODESTRUCT: Code Agents over Structured Action Spaces
arXiv:2604.05407v1 Announce Type: new Abstract: LLM-based code agents treat repositories as unstructured text, applying edits through brittle string matching that frequently fails due to formatting drift or ambiguous patterns. We propose reframing the codebase as a structured action space where...
PaperOrchestra: A Multi-Agent Framework for Automated AI Research Paper Writing
arXiv:2604.05018v1 Announce Type: new Abstract: Synthesizing unstructured research materials into manuscripts is an essential yet under-explored challenge in AI-driven scientific discovery. Existing autonomous writers are rigidly coupled to specific experimental pipelines, and produce superficial literature reviews. We introduce PaperOrchestra, a...
Phase-Associative Memory: Sequence Modeling in Complex Hilbert Space
arXiv:2604.05030v1 Announce Type: new Abstract: We present Phase-Associative Memory (PAM), a recurrent sequence model in which all representations are complex-valued, associations accumulate in a matrix state $S_{t}$ $\in$ $\mathbb{C}^{d \times d}$ via outer products, and retrieval operates through the conjugate...
Context-Agent: Dynamic Discourse Trees for Non-Linear Dialogue
arXiv:2604.05552v1 Announce Type: new Abstract: Large Language Models demonstrate outstanding performance in many language tasks but still face fundamental challenges in managing the non-linear flow of human conversation. The prevalent approach of treating dialogue history as a flat, linear sequence...
TFRBench: A Reasoning Benchmark for Evaluating Forecasting Systems
arXiv:2604.05364v1 Announce Type: new Abstract: We introduce TFRBench, the first benchmark designed to evaluate the reasoning capabilities of forecasting systems. Traditionally, time-series forecasting has been evaluated solely on numerical accuracy, treating foundation models as ``black boxes.'' Unlike existing benchmarks, TFRBench...