{"@attributes":{"version":"2.0"},"channel":{"title":"David Fan","link":"https:\/\/davidfan.io\/","description":"Recent content on David Fan","generator":"Hugo -- gohugo.io","lastBuildDate":"Sun, 27 Feb 2022 00:00:00 +0000","item":[{"title":"Breaking into Industry ML\/AI Research Without a PhD","link":"https:\/\/davidfan.io\/blog\/2022\/02\/breaking-into-industry-ml-ai-research-without-a-phd\/","pubDate":"Sun, 27 Feb 2022 00:00:00 +0000","guid":"https:\/\/davidfan.io\/blog\/2022\/02\/breaking-into-industry-ml-ai-research-without-a-phd\/","description":"I am currently an Applied Scientist who is doing full-time machine learning (ML) research at Amazon without a PhD. I get to work on intellectually difficult problems with strong potential for greenfield innovation, work with really bright and motivated people, and earn high industry pay\u00b9 while doing what I love. Unfortunately, while a lot of people are interested in entering machine learning, there isn\u2019t a lot of guidance online for those trying to transition into ML from software engineering."},{"title":"Paper Summary for ShotCoL: Self-Supervised Video Representation Learning for Scene Boundary Detection in Movies and TV Episodes","link":"https:\/\/davidfan.io\/blog\/2021\/06\/paper-summary-for-shotcol-self-supervised-video-representation-learning-for-scene-boundary-detection-in-movies-and-tv-episodes\/","pubDate":"Sun, 20 Jun 2021 00:00:00 +0000","guid":"https:\/\/davidfan.io\/blog\/2021\/06\/paper-summary-for-shotcol-self-supervised-video-representation-learning-for-scene-boundary-detection-in-movies-and-tv-episodes\/","description":"15 minute read.\nDisclaimer: this paper summary exclusively represents my own views and DOES NOT represent the views of my employer nor my institution. This post should not in any way be treated as an official reference.\nRoughly a year ago, I started doing research in self-supervised video representation learning with a focus on understanding long-form content. That work culminated in a second-author CVPR 2021 paper, which is the focus of this post."},{"title":"Reflections on 2010\u20132019","link":"https:\/\/davidfan.io\/blog\/2020\/03\/reflections-on-20102019\/","pubDate":"Mon, 30 Mar 2020 00:00:00 +0000","guid":"https:\/\/davidfan.io\/blog\/2020\/03\/reflections-on-20102019\/","description":"Time flies! In 2010, I was in 7th grade and just beginning to mature. In 2015, I graduated from high school in NJ and entered college. In mid-2019, I graduated from college and moved from NJ to the West Coast for my first full-time job. To commemorate the ups-and-downs of the past decade, I decided to reflect on my experiences and what I learned from them. Those takeaways are condensed into four themes, in no particular order."},{"title":"The Story of Princeton University Science Olympiad","link":"https:\/\/davidfan.io\/blog\/2018\/12\/the-story-of-princeton-university-science-olympiad\/","pubDate":"Tue, 25 Dec 2018 00:00:00 +0000","guid":"https:\/\/davidfan.io\/blog\/2018\/12\/the-story-of-princeton-university-science-olympiad\/","description":"Science Olympiad is a team-based STEM competition that approximately 8,000 middle school and high school teams compete in nationwide. Science Olympiad was a major part of my life from 7th to 12th grade, and an experience that I am very grateful to have had. In fall 2016 (my sophomore year), I cofounded the Princeton University Science Olympiad invitational tournament with Edison Lee. Co-founding this tournament and scaling it up to become one of Princeton\u2019s largest student-run organizations, has been among the most difficult yet rewarding endeavors I have undertaken."},{"title":"Behind the Scenes: Challenges of Organizing HackPrinceton","link":"https:\/\/davidfan.io\/blog\/2018\/11\/behind-the-scenes-challenges-of-organizing-hackprinceton\/","pubDate":"Sun, 04 Nov 2018 00:00:00 +0000","guid":"https:\/\/davidfan.io\/blog\/2018\/11\/behind-the-scenes-challenges-of-organizing-hackprinceton\/","description":"If you\u2019ve ever attended a hackathon, you know how fun working on projects and learning in a fast-paced environment can be. There\u2019s tons of technical workshops, speakers, mentors, great people, free food, energy drinks, hardware, and fun events. Maybe a bit of sleep too :)\nA lot of work goes behind organizing a large hackathon like HackPrinceton, but not all of the work is \u201ctechy\u201d. We grapple with many interesting logistical, communication, and management challenges, and I hope to give a glimpse of what some of those entail."},{"title":"About Me","link":"https:\/\/davidfan.io\/about\/","pubDate":"Mon, 01 Jan 0001 00:00:00 +0000","guid":"https:\/\/davidfan.io\/about\/","description":"Short Bio I graduated Magna Cum Laude from Princeton University in 2019 with a B.S.E in computer science. I am currently an Applied Scientist at Amazon Prime Video where I do research in video understanding and multimodal representation learning, with an emphasis on self-supervised approaches. I am passionate about advancing the state-of-the-art, publishing my work, and translating research to solve real-world problems. I envision a future where computers are capable of human-like perception and visual reasoning."},{"title":"Contact Me","link":"https:\/\/davidfan.io\/contact\/","pubDate":"Mon, 01 Jan 0001 00:00:00 +0000","guid":"https:\/\/davidfan.io\/contact\/","description":"Feel free to shoot me a message at anytime!\nEmail: dfan (at) davidfan.io\nGitHub: github.com\/dfan\nLinkedIn: linkedin.com\/in\/davidfan97"},{"title":"Experience","link":"https:\/\/davidfan.io\/experience\/","pubDate":"Mon, 01 Jan 0001 00:00:00 +0000","guid":"https:\/\/davidfan.io\/experience\/","description":"Industry Amazon | Seattle, WA → New York, NY\nApplied Scientist, May 2021 - present\nResearch Engineer, July 2020 - May 2021\n Trained 1B param multimodal foundation model with large-scale vision-language-audio pretraining. Out- performs OpenAI CLIP by 25% on internal zero-shot classification and retrieval benchmarks. Enabled automated video advertisement insertion (CEO-level goal) with novel video segmentation model. Developed embeddings for visual search and recommendation which outperform baseline recsys by 5%."},{"title":"Projects","link":"https:\/\/davidfan.io\/projects\/","pubDate":"Mon, 01 Jan 0001 00:00:00 +0000","guid":"https:\/\/davidfan.io\/projects\/","description":"Motion-Guided Masking for Spatiotemporal Representation Learning BERT popularized masked autoencoders for language modeling. Recent works have extended masked autoencoder to image and video domain by reconstructing randomly masked out image\/video patches. These works show that MAE is a promising paradigm for scaling up self-supervised video pretraining. However, random masking assumes that information density is uniformly distributed.\nIn this work, we argue that masking saliency is key to elevating the efficacy of MAE for video."},{"title":"Publications","link":"https:\/\/davidfan.io\/publications\/","pubDate":"Mon, 01 Jan 0001 00:00:00 +0000","guid":"https:\/\/davidfan.io\/publications\/","description":"Peer-Reviewed  Motion-Guided Masking for Spatiotemporal Representation Learning David Fan*, Jue Wang, Shuai Liao, Yi Zhu, Vimal Bhat, Hector Santos-Villalobos, Rohith MV, Xinyu Li.\nICCV 2023.\nPaper, Amazon Blog\n    Nearest-Neighbor Inter-Intra Contrastive Learning from Unlabeled Videos David Fan*, Deyu Yang, Xinyu Li, Vimal Bhat, Rohith MV.\nICLR 2023 Workshop on Mathematical and Empirical Understanding of Foundation Models.\nPaper, [Blog Post Coming Up]\n    Shot Contrastive Self-Supervised Learning for Scene Boundary Detection Shixing Chen, Xiaohan Nie*, David Fan*, Dongqing Zhang, Vimal Bhat, Raffay Hamid."}]}}