Information Technology Representative

+84-091******* Thu Duc, Ho Chi Minh City **************@*****.*** minh1409

MINH LE

LÊ NHẬT MINH

Objective

Wants to cooperate, learn and discover

novelties in the field of artificial intelligence

with talented people for self-improvement

Reference

Assoc. Prof. Duy-Dinh Le

*****@***.***.**

Activity

B2DL-UIT

ELO@UIT K2020

Skill Highlights

• Programming: ML-related Python (JAX, PyTorch, Tensorflow), Statistical Visualization (Matplotlib, Seaborn), C++ for general use, OOP

• Mathematics: Discrete Mathematics (Sets, Graphs, Matrices), Calculus, and Statistical Probability

• CS-related: DL for feature extraction of image, natural language and information structure (set, sequence, graph), dynamic modeling, RL modeling (DQN, AC), AutoML, DSA, Optimization

• Language: Vietnamese, English

Education

2017 – 2020

Le Quy Don High School for the Gifted, Quy Nhon – Binh Dinh I am a former student majoring in Informatics at Le Quy Don High School for the Gifted, Quy Nhon – Binh Dinh. GPA: 8.5/10

2020 – Present

University of Information Technology – VNUHCM

I am currently a junior in the honor program – computer science at University of Information Technology, VNUHCM.

GPA: 8.5/10

Achievements

• Odon Vallet Scholarship (2020)

• Consolation Prize of Vietnam Olympiad in Informatics (2020)

• Second Prize in Informatics Olympiad for University Students (2021) Certificates/Courses

• Problem Solving Using Computational Thinking

• IELTS 7.5

+84-091******* Thu Duc, Ho Chi Minh City **************@*****.*** minh1409 MINH LE

LÊ NHẬT MINH

Projects

Financial Portfolio Generation using Diffusion Method

• Description: A system receives time series market data and the current portfolio and outputs an improved version of the portfolio.

• Proposed method: The method includes formulation of financial portfolios as discrete distributions, an environment returning the difference of reward and risk of the portfolio transition using optimal transport, and MLPs to learn the differences between portfolio and an improved version of that portfolio. The state is encoded using a modified deepset model representing the whole market concatenating with individual stock’s feature vector.

• Gains: An improvement of 0.9% compared to the average annual return using (2007-2013) training data,

(2014-2017) validation data, and (2018-2022) testing data. Tree-based Zero-Cost Proxy Ensemble Initialization for Evolutionary NAS

• Description: A hierarchical method for search space partition combining with non-dominated zero-cost proxy ensemble as evaluating method.

• Proposed method: Certain NAS population-based heuristic solvers can be divided into two phases: initialization phase and search phase. In the initialization phase, the search space is first hierarchically partitioned, then, the performance of each sub-search-space is measured via zero-cost proxies and their representative architectures. The non-dominated sub-search-spaces are selected and partitioned until the search space becomes small enough and selected into initialization.

• Gains: Decrease the search cost of evolutionary NAS threefold compared to random initialization. Interactive Event Retrieval System

• Description: A web-based system to retrieve visual event from videos using natural language.

• Proposed method: The keyframe extraction process utilizes the Ffmpeg framework to extract keyframes from the input media. These keyframes, along with the text query, are then processed by the CLIP model to extract meaningful features. The SCANN model is employed to find the nearest keyframes that match the given query. The web system displays the information of the keyframes, enabling users to select the most appropriate keyframe for each text query efficiently.

• Gains: Knowledge of building machine learning solution

Contact this candidate