Recent:
< Older Updates > :
- 2019/06/24: I just moved back to South Bay area after an exciting and successful stint at BlackThorn Therapeutics. While my team at BlackThorn accomplished a lot and did pioneering work in CNS biotech space, the 2-3 hour each-way commute was really painful to keep up with. So I had to make some changes. My new role is to lead the Applied Research Group in LinkedIn's Data Science organization. It's an honor to join such an esteemed organization that lays claim to the origins of the term "data scientist". I am looking forward to learning a lot, growing professionally, and making a difference.
< Older Updates > :
- 2017/11/06: I've joined BlackThorn Therapeutics as their Head of Data Science. Their mission / vision is amazing. I have a lot to learn about BioTech, and translational medicine in particular. I am looking forward to learning and contributing what I can. Let the "drinking from the fire hose" begin!
- 2017/11/03: I've resigned from Instart Logic, and plan to move to BioTech domain to help build ML algorithms to improve drug discovery and targeting for CNS (central nervous system) diseases like depression, autism and schizophrenia. My time at Instart Logic was an amazing journey and a wonderful learning experience. I am really excited about the opportunities & challenges in biotechnology space that lay ahead.
- 2017/10/28: Our paper on perceived performance got selected into ACM-CCR (Computer Communication Review) journal. It's a nice cherry on the cake!
- 2017/08/21: Our paper won the 2017 ACM-SigComm Internet-QoE best paper award! This paper summarized the insights gleaned from analyzing end-user quality of experience (QoE) of InternetRetailer top-500 websites (Phase-1 of SpeedPerception experiment).
- 2017/07/14: Our EyeQ phase-1 paper on studying progressive image delivery differences between mobile and desktop platforms was presented at 2017 ICME-MMC workshop.
- 2017/06/23: Estelle and I gave well-received talk at SFHTML5 meet-up group at Google SF offices. It was one of the best organized meet-ups I've ever been to (had raffles, speaker prizes and what not). <YouTube video of our talk/Q&A.>
- 2017/04/10: Visited UC-Berkeley to give a seminar at the EECS department. My talk title was: "ML for web application security: Reconciling two competing philosophies".
- 2017/04/04: Our first formal paper summarizing results from SpeedPerception Phase-1 experiment (Chrome / Desktop / IR500) is now publicly available on ArXiv. See: https://arxiv.org/abs/1704.01220. I am particularly excited about the simple ML model that achieves > 85% accuracy in predicting human judgments of perceived speed comparisions between webpage pairs.
- 2017/03/17: Visited UC-Berkeley to give a seminar on perceived web app performance at the Berkeley Artificial Intelligence Research Lab internal seminar series. My talk title was: "Beyond navigation metrics: Teaching computers how users perceive web application performance".
- 2017/03/15: Attended "Strata+Hadoop World 2017" conference (my first time) to present some of our findings on applications of machine learning in security. The talk was surprisingly well recieved. I enjoyed the entire security analytics track and the many conversations with other speakers, and attendees. It was a very good conference experience.
- 2017/03/01: Our awesome developer evangelist Estelle Weyl and I presented the latest results from SpeedPerception study at ForwardJS conference. It was my first time speaking in front of a front-end developer audience.
- 2017/02/23: I was interviewed on O'reilly Data Show Podcast by Ben Lorica. The interview is titled "Building machine learning solutions that can withstand adversarial attacks". We discussed many topics including how we are currently thinking about ML/security landscape.
- 2017/02/14: Spent all day visiting various vendor stalls at RSA-2017. It gave me a good sense the security product landscape, especially in terms of how machine learning is being deployed in production-ready security systems.
- 2017/01/10: We hosted visitors from Telstra for an on-site visit. It was a nice opportunity to show the recent innovations from DS&ML team's work and get some constructive feedback.
- 2016/12/15: The entire DS&ML team and a couple of collaborators went for hotpot / korean bbq lunch in south San Jose. It was fun!
- 2016/11/29: Presented a talk on perceived performance of web applications at LinkedIn offices, and learned about their plans to move to single-page application.
- 2016/10/27: Presented a talk on perceived performance of web applications at SF/Silicon Valley Web Perfornamce Meet-up group. Talk went well, and audience interaction was of good quality. The most fun thing about the talk was the chance to give the talk dressed as a zombie (after our company's zombie social hour earlier that day).
- 2016/10/13: SpeedPerception benchmark, analysis code and initial set of results went public. Everything related to the study is available on GitHub; we want the results to be reproducible, so we made everything accessible.
- 2016/10/06: Gave a talk ( Understanding Complex Web Performance Measurement) at StarWest2016 conference. Focus of the talk was about statistical considerations that go into large-scale performance testing, and how that relates to web performance analysis. I also introduced a new nonparametric method for scenarios where underlying noise is highly variable (to the point where histogram shapes vary wildly). Talk went well and I learned a few things about the current business ecosystem of performance measurement.
- 2016/10/05: We started the process of previewing the first-cut results from Phase-1 of SpeedPerception challenge to important stake-holders in web performance community to get constructive feedback on our findings and gather input about next steps.
- 2016/09/30: Data collection for Phase-1 of SpeedPerception challenge has ended. Phase-1 was a resounding success (more than 75,000 votes cast already - completely on a voluntary basis!). We are finalizing the initial analysis and benchmark for public release.
- 2016/09/09: Attended the beautifully organized workshop on Web App Security at Stanford. Kudos to Dan Boneh, Jas Nagra and Megan Yahya for putting it all together. I had to completely recalibrate my talk based on the ML/security comments during the morning panel and some other chats I had before lunch. It was my first external talk on how we think about web application security from machine learning perspective. The technical quality of the workshop overall was excellent.
- 2016/09/07: We hosted a small team of visitors from Mozilla's FireFox team at Instart Logic offices. If some of our requests (especially related to enabling some additional FireFox features on WebPagetest) accelerate through these idea exchanges, that would be awesome.
- 2016/08/22: FireFox team invited us to come visit so we went to Mozilla offices in Mountain View. Cool digs! Harald Kirschner arranged a very nice lunch session to talk to people, and my post-lunch tech talk about end-user focused web performance measurement found receptive audience. They are also conducting a lot of user studies (which is cool) and they shared some of the recent insights with us. It was a productive visit.
- 2016/08/15: Attended the invite-only Intel Analytics Summit in San Francisco. Pedro Domingos gave a keynote. I enjoyed the panel discussions (the one moderated by O'reilly's Ben Lorica was especially good). The afternoon round-table discussions were under-par.
- 2016/08/02: All of our team went to attend the ScaledML conference at Stanford. A lot of deep learning talks, although it didn't feel like I learned much listening to them (with the exception of Jeff Dean's talk may be). Watching John Canny talk about ML was very interesting; I still hold his original image processing work with very high regard, so the respect remains.
- 2016/07/28: SpeedPerception challenge has been made public. This is a collaborative, open-source, gaming-based effort to create a systematic benchmark of how human end-users perceive popular webpages (in terms of user-experience). If you have 5 minutes to spare, please play and help us move the field forward!
- 2016/07/02: Perceptual Speed Index (a visual performance metric for webpages) and a blog post that explains the rationale behind its design are out in public now (Slide Deck / Blog Post / GitHub Code).
- 2016/06/23: Met WebPageTest's creator Patrick Meenan at Velocity. We talked for a long time on various topics. I love what WebPageTest stands for, and how it enabled me to understand web performance nitty gritty.
- 2016/06/22: Attended the popular web performance conference Velocity in Santa Clara, CA. There was a talk on using machine learning to probe the relationship between web performance and conversion (e-commerce term). It was apparently the first ever ML talk at Velocity.
- 2016/06/14: Jiawen (who did an internship with me in 2015) successfully wrapped up her masters thesis at Imperial College, London. Her topic was directly related to the internship topic I got her started on. I was a co-advisor on her thesis as well, so I felt very happy to see how her internship evolved all the way to a masters thesis. Kudos to her.
- 2016/06/08: Met Paul Irish (Google Chrome) and Shubhie Panicker at Google HQ to present our work on perceptual measurements of webpage loading process. It is a bit funky to meet people (in person) whom you've only seen in online videos before. Excellent idea exchange. Learned some cool things about how Chrome team is looking at measurements of user-experience.
- 2016/05/26: I was invited to attend the "Workshop on the Future of Cyber-Physical Systems" at UC-Berkeley; gave me a chance to catch up with old friends from my alma mater. The event was organized to honor my PhD advisor Shankar Sastry.
- 2016/04/18: I was quoted in Lisa Morgan's article Data Outliers: 10 ways to prevent big data damage in InformationWeek magazine.
- 2016/02/16: I was invited to attend the Netflix Workshop on Personalization, Recommendation and Search (PRS 2016). The conversations were of excellent quality, and the small size of the crowd allowed for a good interaction and idea exchange. Among the talks, my top-2 favorites were by David Ross from YouTube/Google and Olivier Chappelle from Criteo.
- 2015/11/13: Attended MLConf-San Francisco, a business-oriented machine learning conference. It's quite a different setting compared to NIPS / ICML types of machine learning conferences so I found it interesting to sit back and observe. The best talk was probably by Xavier Amatriain (now at Quora, previously at Netflix).
- 2015/08/24: I was an invited panelist for Building/Scaling Your Data Science Team: Best Practices Panel, organized at PagerDuty (SF). Meeting fellow panelists Eric Colson (StitchFix) and Raaid Ahmad (Weebly) in person was a fun experience.
- 2015/07/02: Our technical paper on SmartVision-driven image-rich web application delivery was presented at IEEE International Conference on Multimedia & Expo (ICME) in Torino, Italy. The focus in this paper is more on the systems aspects and QoE gains.
- 2015 /06/15: My long-running collaborative project (with CRG's Mattheiu Louis Lab) from HHMI-Janelia days made it into eLife (open source equivalent of Nature/Science, if I may say so). It's an avant-garde project that was the first to use my OpSIN software to understand olfactory sensory neuron responses by combining behavior, optogenetics and electrophysiology. Ronald Calabrese wrote a very nice primer on our article as well.
- 2014/11/04: Our technical paper on SmartVision algorithm for unsupervised QoE-aware clustering was presented at ACM Multimedia Conference in Orlando, FL, USA. The focus in this paper is more on the core algorithmic aspects.