About
Are you worried about tech debt piling up in your teams' codebases? Are your engineers…
Activity
-
Today is a monumental day for us at PeerDB (YC S23). We are very excited to share that PeerDB will be joining forces with ClickHouse. When we started…
Today is a monumental day for us at PeerDB (YC S23). We are very excited to share that PeerDB will be joining forces with ClickHouse. When we started…
Liked by Yu Hai Lu
-
It was very special to come back to Y Combinator. I gave a panel with Diana Hu, Lucas Ochoa, and William Zeng to the #S24 batch last week and it…
It was very special to come back to Y Combinator. I gave a panel with Diana Hu, Lucas Ochoa, and William Zeng to the #S24 batch last week and it…
Liked by Yu Hai Lu
-
Introducing FlashAttention-3 🚀 It’s optimized for H100, 1.5-2.0x faster than FlashAttention-2 with FP16, up to 740 TFLOPS, i.e., 75% utilization of…
Introducing FlashAttention-3 🚀 It’s optimized for H100, 1.5-2.0x faster than FlashAttention-2 with FP16, up to 740 TFLOPS, i.e., 75% utilization of…
Liked by Yu Hai Lu
Experience
Education
Publications
-
Personifications are Cunning: Exploring Approaches For Personification Identification (Abstract)
Text as Data Analysis 2021
An exploration on approaches for identifying the personification figure of speech in unstructured data such as natural language.
Other authors -
-
Retrieve, Caption, Generate: Visual Grounding for Enhancing Commonsense in Text Generation Models
Association for the Advancement of Artificial Intelligence 2022
We investigate the use of multimodal information contained in images as an effective method for enhancing the commonsense of Transformer models for text generation. We perform experiments using BART and T5 on concept-to-text generation, specifically the task of generative commonsense reasoning, or CommonGen. We call our approach VisCTG: Visually Grounded Concept-to-Text Generation. VisCTG involves captioning images representing appropriate everyday scenarios, and using these captions to enrich…
We investigate the use of multimodal information contained in images as an effective method for enhancing the commonsense of Transformer models for text generation. We perform experiments using BART and T5 on concept-to-text generation, specifically the task of generative commonsense reasoning, or CommonGen. We call our approach VisCTG: Visually Grounded Concept-to-Text Generation. VisCTG involves captioning images representing appropriate everyday scenarios, and using these captions to enrich and steer the generation process. Comprehensive evaluation and analysis demonstrate that VisCTG noticeably improves model performance while successfully addressing several issues of the baseline generations, including poor commonsense, fluency, and specificity.
Other authorsSee publication -
Mastering spaCy (Technical Reviewer)
Packt
- Technical reviewer for the NLP book "Mastering spaCy" by Duygu Altinok
- Annotated, commented and provided feedback for an introductory book on the NLP Python package spaCy
Citation:
Altinok, D. (2021). Mastering Spacy: an end-to-end practical guide to implementing nlp applications using the python... ecosystem. (R. Kumar, T. Shaikh, S. Pandey, S. Editing, K. Lu, U. Yaseen, S. Roy, & C. Schiaffin, Eds.). PACKT PUBLISHING LIMITED. -
PyATE: Python Automated Term Extraction
Zenodo
PyATE is a Python implementation of term extraction algorithms such as C-Value, Basic, Combo Basic, Weirdness and Term Extractor using spaCy POS tagging.
PyATE extracts term candidates from natural languages and assigns each of the candidates a "termhood", a float which can be interpreted as the relative confidence of the candidate being a term.
A distinction needs to be made between "keywords" and "terms". Terms are technical and often specific to a domain and often includes…PyATE is a Python implementation of term extraction algorithms such as C-Value, Basic, Combo Basic, Weirdness and Term Extractor using spaCy POS tagging.
PyATE extracts term candidates from natural languages and assigns each of the candidates a "termhood", a float which can be interpreted as the relative confidence of the candidate being a term.
A distinction needs to be made between "keywords" and "terms". Terms are technical and often specific to a domain and often includes jargon while keywords are mainly for categorizing documents. Formally, “a term is the designation of a defined concept in a special language by a linguistic expression. A term may consist of one or more words." (ISO 1087)
Source code can be found at https://github.com/kevinlu1248/pyate. Documentation can be found at https://kevinlu1248.github.io/pyate/ and a web app demonstrating the algorithms can be found at https://pyate-demo.herokuapp.com/.
Honors & Awards
-
Faculty of Mathematics Entrance Scholarship ($10,000)
University of Waterloo Faculty of Mathematics
A $10,000 merit-based scholarship granted on entrance with an early acceptance, granted to approximately top 10 out of 1000 first-year math students
-
BC Achievement scholarship ($1250)
Government of British Columbia
Merit-based scholarship offered to top 8,000 graduating students in BC annually
https://www2.gov.bc.ca/gov/content/education-training/k-12/support/scholarships/provincial-scholarships/bc-achievement-scholarships -
R.A. Pyke Scholarship for Mathematic recipient ($500)
Sir Winston Churchill Secondary School
-
Chacko and Lize Simon Scholarship ($2,000) [Declined]
Chacko and Lize Simon Scholarship Board
http://www.simonsecret.org/fund.html
-
President’s Scholarship of Distinction ($2,000)
University of Waterloo Faculty of Mathematics
For applying with a high school average of above 95%
Test Scores
-
SAT Math 2
Score: 800
Languages
-
English
Native or bilingual proficiency
-
Chinese
Native or bilingual proficiency
-
Japanese
Limited working proficiency
More activity by Yu Hai
-
Today, we announce Greptile's $4.1M seed round, led by Initialized Capital with participation from Y Combinator, Richard Aberman, and JJ Fliegelman.…
Today, we announce Greptile's $4.1M seed round, led by Initialized Capital with participation from Y Combinator, Richard Aberman, and JJ Fliegelman.…
Liked by Yu Hai Lu
-
AI isn’t going to replace software engineering, but it can make us better. As an engineer, you might have 30 minutes between meetings to review a PR…
AI isn’t going to replace software engineering, but it can make us better. As an engineer, you might have 30 minutes between meetings to review a PR…
Liked by Yu Hai Lu
-
As a SWE intern at Keywords AI (YC W24), one of my projects is testing new features. If I can’t integrate something within a couple of minutes, I…
As a SWE intern at Keywords AI (YC W24), one of my projects is testing new features. If I can’t integrate something within a couple of minutes, I…
Liked by Yu Hai Lu
-
For some time now, Infisical has been publicly available on AWS Marketplace 🛒 If you're interested in simplifying your purchase process of…
For some time now, Infisical has been publicly available on AWS Marketplace 🛒 If you're interested in simplifying your purchase process of…
Liked by Yu Hai Lu
-
I'm back to building! Excited to share I've started a company with Paril Jain and Luke Holoubek. We're building bots that do chores so you don't have…
I'm back to building! Excited to share I've started a company with Paril Jain and Luke Holoubek. We're building bots that do chores so you don't have…
Liked by Yu Hai Lu
-
Every time someone asks me about Devin, I tell them it looks cool, but the most cracked engineers I know are cooking something even better. The cat's…
Every time someone asks me about Devin, I tell them it looks cool, but the most cracked engineers I know are cooking something even better. The cat's…
Liked by Yu Hai Lu
Other similar profiles
Explore collaborative articles
We’re unlocking community knowledge in a new way. Experts add insights directly into each article, started with the help of AI.
Explore More