Hong

Welcome to Hong Hong’s Project Page

About Myself

I am an applied scientist at Meta and previously MSFT, with 9+ years of science modeling and 4+ years of engineering development experiences. Here is my summary:

• Experienced in R&D in areas like LLM, DocAI, NLP, KG, as well as training models within compliant boundary
• Passionate in probabilistic modeling with small data; weak / semi supervision & transfer learning in Enterprise settings
• Hands-on experience in distributed system data processing, online / offline experiments and metrics evaluation
• Used to be the expert of high performance / concurrent / Async development with C++

I am an active member (Moderator) in Chinese AI community on the Clubhouse

I am an Co-Host in Chinese AI podcast EnterAI

The Blogs

1. Random and Fair Red Pockets: A Statistical Approach
   • Shared a common but interesting “stats cookie” problem about how to randomly and fairly split money in Chinese Red-Pockets.
   • Discussed 3 different statistical approaches to split the red pockets money.
   • Analyzed different characteristics and demonstrated the sampling process in python code.
2. Commenting on the O1 Implementation
   • Discussed 2 hypotheses on how would the OpenAI O1 were implemented
   • Hypotheses 1 - using MCTS to perform self CoT training
   • Hypotheses 2 - using self-play to align model’s output to human preference
3. Sampling and Estimation Step-by-Step
4. Generative-model and discriminative Model
5. Perfect Coin
6. Boost Series AdaBoost and GBDT
7. A Story about e

The Lab Projects

1. The Toy Sample for Message-Passing Variation Bayesian Inference
   • Simple implementation demo for message-passing based Bayesian inference
   • Capable of building simple Bayesian graph in Gaussian foamily
   • With detailed documentation and sample code
2. Private Domain Topic Representation Training
   • Solve the topic representation training issues under private domain settings (small corpus size)
   • Joint training between topic representation and token representation
   • BERT/ERL pretrained model based

The Fun Coding Life

1. The Airline Price Optimization from Kaggle
   • Basic solution for airline price optimization challenge with detailed explanation.
2. Risk of Collisions in Fast Rolling Hash Implementation
   • Code samples demonstrate a magic case that cause hash collisions in one of the fast rolling hash implementation

Previous Work Projects

1. Microsoft Viva Topics
   • Smartly discover topic based knowledges in your organization
   • Worked on areas over extraction, conflation as well as ranking
2. Answers @ Microsoft Search
   • Prompt direct answers to your search query
   • Worked on Acronym Answers
3. Microsoft Outlook Online
   • The modern enterprise smart email client
   • Worked on areas over people search, email search
4. Visual C++ STL Library
   • The Microsoft version of C++ standard library
   • Contributed to version 2017, 2015
   • Contributed to C++ co_routine design proposal to language committee
5. C++ Rest SDK
   • The open source Rest SDK for C++
   • Worked as one of the primary project contributor
6. Microsoft Parallel Pattern Library
   • The earlier effort from Microsoft for C++ parallel computing
   • Worked as one of the primary project contributor