Automatic Speech Recognition, Natural Language Understand, Research.
Resume:
CONTACT
LE DUC KHOAN
AI ENGINEER
Su Van Hanh Street, Ward 12,
District 10, Ho Chi Minh City
HCM City University of Technology ********.****@*****.*** Computer Science - Bachelor's Degree.
Talented engineer of the Faculty of Computer Science and Engineering.
Member of the Executive Committee of the Faculty of Computer Science and Engineering and AI Club.
Current GPA: 8.47
TOEIC: 670
Sep 2019 - Present
EDUCATION
Crawl information of Users from Facebook using Scrapy-Splash, Selenium. Analyze and pre-process images and texts crawled from Facebook using Numpy, pandas, pyvi, cv2. Built and used pre-trained model Machine Learning and Deep Learning to detect Facebook Users and determine whether they are brokers using Siamese Network, SSD300, Facenet, SVM, FFNN, and TF-IDF. This project uses information on social networks to predict the information of a customer and recommend suitable products for the customer.
WORKING EXPERIENCE
Group Computer Vision affiliated with Rever Company Jun 2021 - Mar 2022 OLLI Technology JSC Jun 2022 - now
Built, develop, and evaluate End-to-End ASR models such as Wav2vec2.0, HuBERT, and Speech2c using Fairseq, Transformers, HuggingFace, and Pytorch framework. Built a Data Labeling system using Label Studio (Python, ReactJS). The system has supported labeling for approximately 500 hours of audio.
Research Training Strategies for the pre-training and finetuning phases of Wav2vec2.0 and Speech2c. Improve ASR model using Masked-CTC, Intermediate-CTC, CNN, and Attention. Research and develop Automatic Speech Recognition models for Vietnamese Virtual Assistants. I am an enthusiastic graduating student with a profound passion for Artificial Intelligence (AI). As a highly adept self-learner, I can read and understand state-of-the-art knowledge, being the first author of 2 papers and co-author of 1 paper. In my working experience, I support the team to research and develop an ASR module that is used by 20% of users. As the AI landscape expands, I am eagerly anticipating the opportunity to further strengthen my expertise in the field, while completing my Master's Degree by the end of 2024.
linkedin.com/in/khoan-le
https://github.com/Khoan-IT
AI Researcher
Research Study Assistant
Predict the interaction Drug-Protein using KNN.
Recommend sites reaction using DBScan.
This is a scientific research project on drug-protein interaction prediction using GNN. The model will make predictions about the potential for drug-protein interactions and suggest the sites where the reaction occurs so that drug makers can create more effective drugs. Contributions: Knowledge of data structures and algorithms
Stack, Queue, Array, Linked list, Hash, Heap.
Quick Sort, Merge Sort, Bubble Sort, Insertion Sort, Breadth-First-Search, Depth-First-Search. Knowledge of Object-oriented programming
Encapsulation, Inheritance, Polymorphism, Abstraction. Knowledge of Artificial intelligence
Model Machine Learning: Linear Regression, Logistic Regression, KNN, K-means, DBScan, SVM, Decision Tree, Naive Bayes, Random Forest, Multi-Layer Perceptron. Model NLP: RNN, LSTM, GRU, Seq2Seq, Transformers, BERT, JointBERT, JointIDSF. Model Speech processing: Wav2vec2.0, HuBERT, Speech2c. Feature Extraction: Fisher’s Linear Discriminant Analysis, Principal Component Analysis. Optimizer NN: Gradient Descent, SGD, Momentum, RMSprop, Adam. Activation Function: Sigmoid, Tanh, ReLU, Leaky ReLU, Softmax. Word Embedding: TF-IDF, CBOW, Skip-gram.
Loss Function: RMSE, Multi-class Cross Entropy.
Frameworks Deep learning: Tensorflow, Pytorch, Keras.
(SEATUC 2023) Towards an Improvement of Slot Filling Method Based on JointBERT.
(NICS 2022) Towards De Novo Drug Design for the Coronavirus: A Drug-Target Interaction Prediction Approach using Atom-enhanced Graph Neural Network with Multi-hop Gating Mechanism.
(SJSE2023-1) Towards an Improvement of Slot Filling Method Using Convolution Layer to Capture Localization
(Under review).
Research Study Assistant
RESEARCH PAPER
BASIC KNOWLEDGE
THESIS
Built and develop Intent Classifications and Slot fillings model to get the purpose as well as extract information in User's questions and posts crawled from the social pages. Proposed a new architecture that combines the strengths of CNN and BERT, significantly improved performance compared with JointBERT and JoinIDSF.
Built an Action decider model to take action based on the state of the current dialogue by using Reinforcement learning techniques.
Took advantage of ChatGPT to generate synthetic data for low-resource scenarios by creating prompts. We build and deploy a Chatbot to assist students in finding information about extracurricular activities at Ho Chi Minh City University of Technology.
Scientific research project on Deep learning Oct 2021 - Jul 2022
Contact this candidate