A5054796820- Recommender Systems and Techniques
- Topic Modeling
- Speech and dialogue systems
- Data Visualization and Analytics
- Complex Network Analysis Techniques
- UAV Applications and Optimization
- Video Analysis and Summarization
- Education and Critical Thinking Development
- Machine Learning in Healthcare
- Reinforcement Learning in Robotics
- Artificial Intelligence in Games
- Advanced Text Analysis Techniques
- Digital Storytelling and Education
- Advanced Bandit Algorithms Research
- AI-based Problem Solving and Planning
- Internet of Things and AI
- Advanced Graph Neural Networks
- IoT-based Smart Home Systems
- Robotics and Automated Systems
- Language, Metaphor, and Cognition
- COVID-19 epidemiological studies
- Time Series Analysis and Forecasting
- Anomaly Detection Techniques and Applications
- Web Data Mining and Analysis
- Educational Games and Gamification
University of California, San Diego
2024
Korea Advanced Institute of Science and Technology
2019-2022
Hypergraphs provide a natural way of representing group relations, whose complexity motivates an extensive array prior work to adopt some form abstraction and simplification higher-order interactions. However, the following question has yet be addressed: How much interactions is sufficient in solving hypergraph task, how different such results become across datasets? This question, if properly answered, provides useful engineering guideline on trade off between accuracy downstream task. To...
Graphs have been utilized as a powerful tool to model pairwise relationships between people or objects. Such structure is special type of broader concept referred hypergraph, in which each hyperedge may consist an arbitrary number nodes, rather than just two. A large real-world datasets are this form - for example, lists recipients emails sent from organization, users participating discussion thread subject labels tagged online question. However, due complex representations and lack adequate...
With the vast and dynamic user-generated content on Roblox, creating effective game recommendations requires a deep understanding of content. Traditional recommendation models struggle with inconsistent sparse nature text features such as titles descriptions. Recent advancements in large language (LLMs) offer opportunities to enhance systems by analyzing in-game data. This paper addresses two challenges: generating high-quality, structured for games without extensive human annotation,...
Calibrated recommendation, which aims to maintain personalized proportions of categories within recommendations, is crucial in practical scenarios since it enhances user satisfaction by reflecting diverse interests. However, achieving calibration a sequential setting (i.e., calibrated recommendation) challenging due the need adapt users' evolving preferences. Previous methods typically leverage reranking algorithms calibrate recommendations after training model without considering effect and...
We consider the Markov Decision Process (MDP) of selecting a subset items at each step, termed Select-MDP (S-MDP). The large state and action spaces S-MDPs make them intractable to solve with typical reinforcement learning (RL) algorithms especially when number is huge. In this paper, we present deep RL algorithm issue by adopting following key ideas. First, convert original S-MDP into an Iterative (IS-MDP), which equivalent in terms optimal actions. IS-MDP decomposes joint K simultaneously...
Given a sequence of epidemic events, can single model capture its dynamics during the entire period? How should we divide into segments to better dynamics? Throughout human history, infectious diseases (e.g., Black Death and COVID-19) have been serious threats. Consequently, understanding forecasting evolving patterns events are critical for prevention decision making. To this end, models based on ordinary differential equations (ODEs), which effectively describe dynamic systems in many...
Synthetic users are cost-effective proxies for real in the evaluation of conversational recommender systems. Large language models show promise simulating human-like behavior, raising question their ability to represent a diverse population users. We introduce new protocol measure degree which can accurately emulate human behavior recommendation. This is comprised five tasks, each designed evaluate key property that synthetic user should exhibit: choosing items talk about, expressing binary...
We introduce a multimodal dataset where users express preferences through images. These images encompass broad spectrum of visual expressions ranging from landscapes to artistic depictions. Users request recommendations for books or music that evoke similar feelings those captured in the images, and are endorsed by community upvotes. This supports two recommendation tasks: title generation multiple-choice selection. Our experiments with large foundation models reveal their limitations these...
In this paper, we present a systematic effort to design, evaluate, and implement realistic conversational recommender system (CRS). The objective of our is allow users input free-form text request recommendations, then receive list relevant diverse items. While previous work on synthetic queries augments large language models (LLMs) with 1-3 tools, argue that more extensive toolbox necessary effectively handle real user requests. As such, propose novel approach equips LLMs over 10 providing...
본 논문에서는 인공지능을 활용하여 드론(무인 항공기)이 목표물을 추적하는 시스템의 설계, 구현, 성능 평가를 소개한다. 목표물 추적하기 위해서는 드론에서 촬영한 영상에서 인지해야 하며 이를 기반으로 드론의 움직임을 결정해야 한다. 인공지능 작업이 요구되는 곳은 정확한 인지를 위해서 신경망 기반의 알고리즘의 적용과 불규칙적으로 움직이는 놓치지 않으면서 동시에 에너지 소모를 최소화 하는 결정하기 강화학습에 기반한 적용이다. 사용자의 명령 전달 및 추적 결과를 확인하기 위하여 지상관제센터(GCS)와 드론간의 네트워크 연결 또한 필수적인 요소이다. 이러한 작업들을 효과적으로 수행하기 위한 플랫폼이 필요하며, 드론 플랫폼은 네트워킹, 연산, 강화학습을 통한 조종기능 지원, 다양성 지원을 제공해야 이런 기능을 요구사항을 정리하였으며, 추적을 기반 플랫폼을 제시하고, 실험과 시뮬레이션을 통하여 검증하였다.
무인항공기(Unmanned Aerial Vehicle 또는 드론)에서 목표물 추적을 위한 2가지 대표적인 인공지능 작업은 (i) 인식 및 추적과 (ii) 드론의 제어이다. 이와 같은 작업들은 높은 실시간 계산을 요구하나 중소형 무인항공기들에서 탑재 가능한 컴퓨팅 능력 한계로 인한 한계가 존재한다. 이를 해결하기 위해 클라우드로 작업을 오프로딩하는 방법이 존재하나, 이는 지연시간을 동반하게 되는 단점이 있다. 본 논문에서는 능력의 존재하는 순수 드론위에서의 작업 수행과, 지연시간이 클라우드 기반의 수행의 성능에 대해 무인항공기 기반 추적 응용에서 비교 분석을 수행한다. 결과를 통하여 현재 기술로 달성할 수 있는 정량화 된 성능과 앞으로 출현하게 될 서비스들의 원활한 수행을 위해서 컴퓨팅능력, 통신망 성능들의 요구사항을 확보할