{"id":51,"date":"2021-05-06T15:44:40","date_gmt":"2021-05-06T15:44:40","guid":{"rendered":"https:\/\/www.let-all.com\/blog\/?p=51"},"modified":"2021-05-06T15:44:58","modified_gmt":"2021-05-06T15:44:58","slug":"alt-highlights-a-report-on-the-first-alt-mentoring-workshop","status":"publish","type":"post","link":"https:\/\/www.let-all.com\/blog\/2021\/05\/06\/alt-highlights-a-report-on-the-first-alt-mentoring-workshop\/","title":{"rendered":"ALT Highlights \u2013 A Report on the First ALT Mentoring Workshop"},"content":{"rendered":"\n

Welcome to ALT Highlights, a series of blog posts spotlighting various happenings at the recent conference ALT 2021<\/a>, including plenary talks, tutorials, trends in learning theory, and more! To reach a broad audience, the series will be disseminated as guest posts on different blogs in machine learning and theoretical computer science. This initiative is organized by the Learning Theory Alliance<\/a>, and overseen by Gautam Kamath<\/a>. All posts in ALT Highlights are indexed on the official Learning Theory Alliance blog<\/a>.<\/p>\n\n\n\n

This is the fifth post in the series, coverage of the first ALT Mentoring Workshop<\/a> organized by the Learning Theory Alliance, written by Keziah Naggita<\/a> and Sutanu Gayen<\/a>.<\/p>\n\n\n\n

\n\n\n\n

1 Introduction<\/strong><\/h2>\n\n\n\n

The Learning Theory Alliance (Let-All) is an online initiative aimed at developing a supportive learning theory community, founded by (1) Surbhi Goel<\/a>, a postdoctoral researcher at Microsoft Research New York, (2) Nika Haghtalab<\/a>, an assistant professor at UC Berkeley EECS, (3) and Ellen Vitercik<\/a>, a Ph.D. Student at CMU; and advised by Peter Bartlett<\/a>, Avrim Blum<\/a>, Stefanie Jegelka<\/a>, Po-Ling Loh<\/a>, and Jenn Wortman Vaughan<\/a>. The goal of the alliance is to ensure healthy community growth by fostering inclusive community engagement and encouraging active contributions from researchers at all stages of their careers. Let-All\u2019s efforts towards realizing these goals include a series of ongoing and future activities, such as the first ALT mentoring workshop, coordinating the ALT Highlights blog series, and other upcoming community initiatives. This article reports on\u00a0 Let-All\u2019s first Mentoring Workshop<\/a>, which was affiliated with the 32nd<\/sup> International Conference on Algorithmic Learning Theory.<\/p>\n\n\n\n

The workshop had two main sessions to cater to the time zone differences of the participants.  These sessions had three main components: an academic program, which included how-to-talks, Ask Me Anythings (AMAs), and presentation dissections; a technical program, which included research talks; and a social program, which included discussion tables and other activities.<\/p>\n\n\n\n

The workshop participants included students, researchers, and industry professionals, all at different levels of familiarity with learning theory. Because of the ongoing COVID-19 pandemic, the workshop was virtual. It was held on the online platforms Zoom and Gather town, a virtual interactive environment that mimics an in-person workshop setting. For accessibility, the workshop organizers opened up the workshop free of cost to all registered participants. <\/p>\n\n\n\n

2 Program Highlights<\/strong><\/h2>\n\n\n\n

2.1 Academic Program<\/strong><\/h3>\n\n\n\n

To kick off the workshop, one of the organizers began with a welcome lecture: Surbhi in session one and Nika in session two.  They read out the code of conduct and who to contact in case of issues, outlined the workshop\u2019s purpose, and gave attendees demographic information. They explained how participants could navigate the workshop-themed Gather town workspace and then ended the introduction with encouragement for participants to mingle. <\/p>\n\n\n\n

The How-to-Talks<\/em><\/strong> sessions covered writing papers, giving talks, and networking. In Session 1, Pravesh Kothari<\/a> talked in great detail about the dos and don\u2019ts of what to add in the abstract, overview, introduction, and appendix when advising participants on how to best structure research papers. He told attendees to always put effort into understanding their intended reader or talk audience.  Pravesh encouraged attendees to consider the expertise and interests of the reader or listener to capture their attention since these highly determine the attention span and interest in the information presented to them.  He strongly recommended attendees watch the Leadership Lab: The Craft of Writing Effectively<\/a><\/em> by Larry McEnerney , Director of the University of Chicago Writing Program. <\/em>In session 2 of the workshop, Rafael Frongillo<\/a>, similar to Pravesh, discussed how to capture the intended audience when one writes a paper, reviews, and talks.<\/p>\n\n\n\n

In the first networking session<\/em><\/strong>, Jacob Abernethy<\/a> encouraged participants to seek out horizontal and vertical networking, for example, through collaborations, talks, and reach outs. He said that currently, in academia, Ph.D. admissions, faculty hiring, and tenure appointments are heavily risk-averse. Therefore, people seek out candidates based on their network. For this reason, it is crucial for students to network from early on in their careers. He gave great examples of how junior researchers can reach out and forge relationships with other researchers. For example, when you meet academics, faculty\/postdocs at events, ask to give a talk at their lab. Jacob also candidly talked about his earlier failures at MIT and how they shaped his journey. He talked about luck and how John Langford<\/a>, who was at Toyota Technological Institute at Chicago at the time, took a chance on him that forever changed his life. Jacob, therefore, advised academics to take chances on people as this would change the course of the field. <\/p>\n\n\n\n

Jamie Morgenstern<\/a> discussed different networking methods in the second How-to-talks session<\/em><\/strong>. She emphasized that for junior researchers, it\u2019s important to attend conferences and to network with others, to advertise their research through talks, and to reach out to faculty for collaboration. To introduce oneself and capture the listener\u2019s attention, Jamie said, for conferences, prepare to do so in two minutes, for social four minutes, bar 12 minutes, and faculty interview 25 minutes. Senior grad students may help introduce the juniors during lunch\/poster sessions. Finally, when emailing faculty about research, she said one should avoid discussion about other people\u2019s work and instead should stick to the recipient\u2019s work \u2013 \u201cshowing deep understanding and possibly open questions which might lead to collaboration.\u201d <\/p>\n\n\n\n

In both workshop sessions, there were two parallel talk dissections, <\/em><\/strong>in which senior faculty members gave both positive and constructive feedback on talks junior researchers presented. In the first session, Bobby Kleinberg<\/a> discussed Emily Diana<\/a>‘s talk titled “Minimax and Lexicographically Fair Learning: Algorithms, Experiments, and Generalization”. He highlighted parts that were impressive, those that needed improvement, and gave general advice on structuring an audience-based presentation. When Bobby suggested including more diagrams than text, a few people made suggestions of free tools including tikz, matcha.io, PowerPoint, and draw.io. In parallel, Kamalika Chaudhuri<\/a> dissected a talk on \u201cEfficient, Noise-Tolerant, and Private Learning via Boosting\u201d by Marco Carmosino<\/a>. Two main takeaways of this talk dissection were the balance of technical and nontechnical content (e.g., explaining ideas with fun pictures, etc.) and having one main and clear idea as the talk\u2019s takeaway. <\/p>\n\n\n\n

In the second session, Mingda Qiao<\/a> gave a talk titled: \u201cStronger Calibration Lower Bounds via Sidestepping\u201d which Praneeth Netrapalli<\/a> dissected. Praneeth remarked that theory folks often jump into the problem straight away without covering much background. In conferences, this might be fine due to time pressure and specific interests. However, in broader settings such as departmental seminars, he advised the speaker to allocate more time to introduce the problem lucidly and concisely. In parallel, Mary Wootters<\/a> dissected a talk titled \u201cList-Decodable Subspace Recovery: Dimension Independent Error in Polynomial Time\u201d that Ainesh Bakshi<\/a> presented. <\/p>\n\n\n\n

In the first AMA session<\/em><\/strong> moderated by Aaditya Ramdas<\/a>, Lester Mackey<\/a> refreshingly answered several of the attendees’ well-curated questions about what makes strong collaborations, how to get into grad school, and whether or not he ever felt like quitting his Ph.D., among others.  He encouraged students to take classes with professors they are interested in as it makes it easy to ask for a mentorship opportunity. Lester talked about collaborations and imposter syndrome and encouraged attendees to look on the brighter side of things, to remember that we all are working towards one big goal, creating positive changes in the world. Therefore if someone discovers a result before us, we should applaud them, collaborate if possible, and move onto new problems. He said he did not necessarily plan to do a Ph.D. but got into it towards the end of his undergraduate degree due to an internship that made him fall in love with doing research. <\/p>\n\n\n\n

In the evening, there was an AMA session with Shafi Goldwasser<\/a> moderated by Nika. Shafi gave thoughtful and candid answers to attendees’ captivating questions about research, life in academia, collaborations, among others. Shafi told attendees that healthy competition, trust, and overlap of research interest, is crucial for successful research in the early stage of the career. She also asserted that fundamental science is always impactful. She mentioned that the high points of her career were working on problems she was curious about: cryptography, pseudo-randomness, and zero-knowledge proofs. Finally, when asked about what advice she wished she had during the early stage of her career, interestingly Shafi replied: \u201chaving good colleagues, good friends at work, very important, most important \u2013 having a listening, promoting and supportive cohort of friends rather than an individualistic path as a scholar is priceless.\u201d <\/p>\n\n\n\n

2.2 Technical Program<\/strong><\/h3>\n\n\n\n

Two research talks<\/em><\/strong> happened in the first session. First, Po-Ling Loh gave a talk titled \u201cMean estimation for entangled single-sample distributions.\u201d Then Vatsal Sharan<\/a> talked about \u201cSample Amplification: Increasing Dataset Size even when Learning is Impossible\u201d.
Similarly in the second session, Nadav Cohen<\/a> gave the first talk about tensor and matrix completion problems and the importance of understanding the theory behind deep learning from theoretical and practical perspectives. After, Zhiyi Huang<\/a> gave a talk titled \u201cSetting the Sample Complexity of Single-parameter Revenue Maximization.\u201d <\/p>\n\n\n\n

2.3 Social Program<\/strong><\/h3>\n\n\n\n

In both sessions, during the social hours, Sumegha Garg<\/a>, Suriya Gunasekar<\/a>, and Thodoris Lykouris<\/a> organized the table topics<\/em><\/strong> to help attendees meet and interact with senior researchers and professors on different topics. The table topics included the following; starting on ML research, research agendas, ML+X: multidisciplinary research, advisor-advisee relationships, collaborators, communicating research and networking, beyond your institution: internships and research visits, planning after grad school: academia versus industry, Grad school applications, Work ethics, and Open research discussion. <\/p>\n\n\n\n

The table topics were chaired by; Jacob Abernethy, Shivani Agarwal<\/a>, Eric Balkanski<\/a>, Peter Bartlett, Avrim Blum, S\u00e9bastien Bubeck<\/a>, Kamalika Chaudhuri, Nadav Cohen, Sumegha Garg, Surbhi Goel, Suriya Gunasekar, Nika Haghtalab, Daniel Hsu<\/a>, Prateek Jain<\/a>, Mike Jordan<\/a>, Sham Kakade<\/a>, Adam Kalai<\/a>, Pravesh Kothari, Akshay Krishnamurthy<\/a>, Jerry Li<\/a>, Po-Ling Loh, Thodoris Lykouris, Yishay Mansour<\/a>, Pasin Manurangsi<\/a>, Vidya Muthukumar<\/a>, Praneeth Netrapalli, Wen Sun<\/a>, Bo Waggoner<\/a>, Manolis Zampetakis<\/a>, and Cyril Zhang<\/a>.\u00a0<\/p>\n\n\n\n

Lastly, at the end of the two general research talks in sessions one and two of the workshop, attendees assembled on Gather town to close the workshop. The social event included 1:1 social interactions with other attendees, an attempt at forging relationships, and activities like dancing.<\/p>\n\n\n\n

3 Attendance Statistics, Testimonials and Feedback<\/strong><\/h2>\n\n\n\n

3.1 Participants Statistics<\/strong><\/h3>\n\n\n\n

ALT mentoring workshop welcomed talented academics, researchers, and professionals from a wide array of backgrounds. Of the 438 registered to attend the workshop, 197 were new to the learning theory community, 37 attended at least one ALT\/COLT conference in the past, and 146 hadn\u2019t attended ALT\/COLT but had attended machine learning conferences (STOC, NeurIPS, etc.) as shown in Figure 1. <\/p>\n\n\n\n

The workshop participants came from different parts of the world, were of different genders, races, and seniority levels. We use Figures 2, 3b, 4a, and 4b to highlight this demographic information about the participants. Some participants chose session one, and others chose session two, a choice driven by their schedules and time zones. The attendance composition is as shown in figure 3a.<\/p>\n\n\n\n

\"\"
Figure 1: Registrants familiarity with the community<\/figcaption><\/figure><\/div>\n\n\n\n
\"\"
Figure 2: Career stages of participants<\/figcaption><\/figure><\/div>\n\n\n\n