Shawndra Hill’s Post

ML in academia vs industry I didn't have a typical path to big tech. My undergrad was in biomedical engineering, my master's in information and computer engineering and my PhD in statistics and computational biology. In my third year of PhD I did a software engineering internship at Bloomberg and joined the Search & NLP group there as soon as I submitted my thesis. I've now been at Meta for the last five years, working on recommender systems and ads delivery optimization. And I've never looked back. I always knew academia wasn't right for me. I wanted to solve real-life problems and build things. Unashamedly, I was also interested in making money and I did not see a way of making that happen in academia. Lastly, I liked to code, more than reading or writing papers. In my head, that’s what I would do *all the time* as a software/ML engineer in industry (plot twist coming up) while being paid a lot more. What was there not to like? The transition was a bit of culture shock. Back in academia I was used to working on my own and being my own bottleneck. I got to do *deep* work for hours/days on end.  That is rarely the case in industry. It’s still of course about solving difficult technical problems, but it’s also about navigating complex human dynamics, figuring out how to work in a team, how to align with others with different agendas and how to manage resources, timelines and competing priorities. The problem complexity is also different. In industry, I rarely *have* to read papers because I’ll usually know how to solve a problem from a modeling perspective. The complexity is about how to get our production systems to do that ‘simple’ thing we need. It’s about the complex interaction of components that are maintained and developed by thousands of engineers serving billions of users at a time. It is about your experiment getting corrupted by a team on the other side of the world making a change to a service you know nothing about. Nothing in academia prepares you for this challenge. And it's pretty fun. If you are interested in making the jump from academia to industry check out this job opening at Meta!

#hiring UXR Research Scientist, New York City, United States, fulltime #jobs #jobseekers #careers #NewYorkCityjobs #NewYorkjobs #ITCommunications Apply: https://lnkd.in/gh7MaNsR As a Research Scientist, you are expected to have breadth and depth of scientific background in human perception and cognition. You will be leveraging this background in service of defining, assessing and enhancing the form, function and dynamics of our hardware and software. Using expert data analytics and to develop metrics for understanding and measuring novel challenges in areas including sensors, displays, wearable device form factors. Apply your knowledge of human perception, visual and auditory, to create compelling holistic product experiences. UXR Research Scientist Responsibilities: demonstrate your ability to break down ambiguous problems and master research methods, including but, not limited to, design of experiments, expert review, psychophysics & threshold detection, and relevant comfort research approaches.Create and drive actionable research insights to product management and engineering teams.work across multi-disciplinary hardware and software teams to push the boundaries of RL products while enhancing the value users get from these productsDrive measurable positive impact of the user experience of our products. Minimum Qualifications: Bachelor's degree in Computer Science, Computer Engineering, relevant technical field, or equivalent practical experience.PhD in Cognitive/Experimental Psychology, Cognitive Science, Kinesiology, Biomedical Engineering, Industrial Engineering, Human-Computer Interaction or a closely related scientific or engineering field.Demonstrated scientific background in human multi-sensory perception, cognition, motor control and movement, the experience applying such knowledge in analyzing and interpreting relevant human behaviors associated with novel systems5+ years of experience in experimental design, psychophysics, behavioral analytics, and proficiency with inferential statisticsProven track record of rigorous scientific research, including development of metrics for psychophysical and empirical behavioral measurementsDemonstrated communication skills as well as proven interpersonal skill 2+ years of experience in conducting research for novel products headed to market Preferred Qualifications: Training and experience in related techniques such as eye tracking, body motion tracking, postural stability measures, EMG and other biometric sensing technology, and physiology measuresExperience in research for emerging technologies, AR/VR/MR or wearable technologyExperience in research across full product life-cycleExperience in competing research on visual or auditory perception and consumer products. About Meta: Meta builds technologies that help people connect, find communities, and grow businesses. When Facebook launched in 2004, it changed the way people co

Meta AI is on track to become the most used AI assistant of the world by the end of this year, with hundreds of millions of users using it now. We just upgraded Meta AI’s underlying LLM to Llama 3.1 with much improved reasoning, multilinguality, tool usage and many other aspects. More importantly, our most advanced and largest open-source model Llama 405B is available on WhatsApp and Meta.ai. It brings much improved reasoning capabilities that make it possible for Meta AI to understand and answer your more complex questions. More details: Llama 3.1: https://lnkd.in/gU5Kr2nY Meta AI: https://lnkd.in/g9v_7W4p If the building the most used and advanced AI sounds exciting to you, we are indeed #hiring LLM LLM leaders (multiple locations in the U.S.) to help drive our post-training and agents efforts. Please see below for the job links: Senior Research scientist manager: https://lnkd.in/gZsjv9Sw Research scientist manager: https://lnkd.in/gqUg-p-Z Research scientist lead: https://lnkd.in/gtAXfAJJ

4 of 15 #intern #openings on 12/27 ✴️ Gen AI ✴️ #ml Facebook ✴️#phd #student #checkitout and #share #opportunity #usa #usjobs #hiring #hiringnow #open #job #internship #reshare #engineeringjobs #engineering #givingback #spreadtheword #connectandgrow

https://lnkd.in/dMj25Rym Connecting the dots between the Google AI diaspora & what sounds very similar coming out of Meta looks like a promising job for hiring managers, retained search firms & recruiters. What's the secret sauce driving exceptional talent to bail on both companies? Also worth thinking about: The Information story reports substantial internal competition between Meta silos over genAI infrastructure. Other Operational Risk Managers (focusing on technology) might want to track requests for the same infrastructure at their own businesses. A new KRI? #meta

Here’s how I failed my Machine Learning interview at Meta. One day, while scrolling through LinkedIn, I received an email from a recruiter at Meta. He mentioned that he was impressed with my Covid-19 ML research and would love for me to interview with Meta for the role of ML Researcher. I couldn’t believe it. I was a mechanical engineer just 2 years back, and never ever dreamt of working for a big software giant like Meta. I also thought those roles were only for CS students, and not for mechanical engineers like me! Immediately I replied saying that I was interested. The interview was fixed on a date 1 month later. When I read what was asked in ML interviews, I came across the following: - Core questions about the ML field: I was confident about this. - Detailed explanation of how ML algorithms work: I was confident about this. - Actual implementation of ML algorithms from scratch: I was confident about this. - 2 live coding DSA rounds: I was really scared of this. Upon further reading, I came to know about Leetcode. I had 1 month to practice Leetcode questions and be ready for a live coding round with a Meta engineer. The thought of live coding in front of an experienced, world-class engineer was very, very scary. This was all new to me. Few friends suggested me to practice the 75 most popular Leetcode questions, since only 1 month was left. Every day, for 5-6 hours I practiced Leetcode questions. Easy, Medium and Difficult. In 30 days, I became good at solving the Easy questions and some Medium questions. I could never solve a single Difficult Leetcode question entirely on my own. On the interview day, I just hoped that the questions were easy or medium. Round 1: Longest String with Repeated characters. I solved this successfully. Round 2: Merge k sorted lists. Failed here miserably. They decided to give me a 3rd round, which was the decider round. Round 3: Validate IP address. Got stuck here and couldn’t complete it. They ultimately rejected me. I felt dejected and thought I had let go of a big opportunity. Looking back, it was probably better that I didn’t get in. If I would have solved the 2nd coding question correctly, I would have been working a 9-6 corporate job, for someone else’s vision. Lesson learnt: Even if you don’t get into the flashy, big corporations or universities, life has other better plans for you.

Meta's Llama 3 dropped just a few days ago, but already there are close to 1,000 variants publicly available on Hugging Face. Most people read base model evaluations and think they apply to all these deployments, but that ain't the case. Or at least the extent to which it's true is poorly understood. I want to see more eval work focus on the performance and risk shadow cast by base models on their downstream progeny. Basically, if base models aren't being used as-is, do their evals matter? Or are they poorly predictive of what will come of them post-fine-tuning? I don't have time/resources to work on this right now, but I'd love to see someone take this project on: across all variants posted on HF, how do evals on fine-tuned versions compare to evals done on the original base model? What kind of relationship is there between the two? Is the latter a constraint or directional anchor for the former? The best set-up here needs to find some way of ignoring some amount of noise (e.g., devs creating their own copy but not doing any fine tuning or doing some kind of terrible fine tuning) in order to focus on more serious attempts to fine tune for stronger performance on specific use cases, stronger safety and security, or something else. This question is increasingly important because fine-tuning is becoming easier and faster to do. That means it's more and more likely that developers will tune models to their own liking. Evals done on prior upstream versions are potentially meaningless... 'potentially' is the key word because we don't have a good understanding of the persistence of performance and risk, the elasticity of learning. By the way, if you're interested in working on hard questions around LLM evaluation, like these ones, we're hiring! Our Trust & Governance research team is looking for a Staff Applied Research Scientist. Come help us tackle these challenges! Link to the job posting in the comments. #artificialintelligence #trustworthyai #llm

My team and my org at Meta are hiring software engineers / research scientists w/ machine learning background, of IC5/6. The positions are based in Bellevue, WA. If you or your connections are interested, please ping me or leave a comment. More details are as below. Team Name: Ads Core ML - Signal Growth Ranking Team Pillar: Monetization   Team/Role: ⭐️⭐️⭐️⭐️⭐️Grow Meta's revenue💰, 🚀boost your career with us! ⭐️⭐️⭐️⭐️⭐️   What is the main revenue driver for Meta? Ads. It contributes 98% of the revenue (according to most recent earning report of Meta).   What is the biggest challenge of Ads? Signal loss.   We are a team to solve signal loss problem thru privacy enhanced technology to defend and grow Meta's revenue. We are hiring!   In 3-5 years, we will have enhanced our Ads Ranking ML technologies to be flexible in the face of adhoc privacy regulations at scale while still performant for our advertiser customers and respectful of users' privacy choices. Our team will be responsible for transitioning our ML models from a world of fully supervised learning with access to user-level data to ML techniques that maximize value for our users and customers on top of privacy enhancing technologies.   List a few new initiatives and directions in this area: leveraging LLM in signal growth ranking models, exploring widely adopted DP-ML techniques (i.e. DP-SGD and LabelDP) by the industry, building and enhancing a privacy evaluation system to serve as a safeguard aligning with Meta's dedication to upholding user privacy, leveraging statistical analysis to reduce inconsistency between offline and online metrics. etc. 

Oh the hard answer to this hard question has been sitting here in plain sight. How do we evaluate discrete points on an exponential by use case? The answer has been in the realm of image, the world of Stable Diffusion. A billion models easily merged and tuned. Most useless. Some overused. None specific or perfect for any particular thing. Each a labor of love by an individual among thousands in a fractal community of builders. Behold the beauty of the untamed open source ecosystem. We will never sort and compartmentalize it, and it will grow exponentially, maddeningly, unhindered by the forces of order and alignment. We no longer need to worry about which model or even which fine-tune is the best. The free market has finally arrived, and with it a million builders. Thank you Yann LeCun , OG liberator of models. #opensource

Summary: Meta's chief AI scientist, Yann LeCun, revealed that he turned down a job offer for director of research at Google in 2002 due to factors including low compensation and the company's size and revenue at the time. He also mentioned that his decision was influenced by the need to focus on research and his family's reluctance to move to California. Takeaway: Yann LeCun's decision sheds light on the importance of considering various factors beyond financial compensation when evaluating job opportunities, such as the alignment with personal and professional goals, and the potential impact on company culture and research direction. Hashtags: #AI #MachineLearning #Meta #Google #YannLeCun #CareerDecisions