So incredibly honored to have received this recognition for our work! Thanks to the #sigir2020 program committee & the reviewers. Huge thanks to my amazing team--Marco Morik, Jessica Hong, and especially my advisor, Thorsten Joachims. Couldn't have asked for a better team!

Excited to announce 2020 ACM SIGKDD Innovation Award is given to Thorsten Joachims (Cornell University) for his research contributions in machine learning, including influential work studying human biases in information retrieval, SVM, and structured output prediction.

In Information Retrieval, we've had a line of work focusing on mechanisms to distribute attention in rankings: dl.acm.org/doi/10.1145/32… dl.acm.org/doi/10.1145/32… These model how people consume what's displayed. This thread is a great summary of other factors at play.

☝ We have a recommendation for that Introducing TensorFlow Recommenders, an open-source package that makes building, evaluating, and serving recommender models easy. Find recommendations for movies, restaurants, and much more! Get started → goo.gle/2HrCwDq

I wrote a blog post on how baselines affect much more than variance in policy gradient methods. It discusses a paper we recently put out on arXiv. Blog post: mcmachado.info/?p=328 Paper: arxiv.org/abs/2008.13773 This paper is with Wes Chung, @lanternol, and @le_roux_nicolas.

I'm delighted to share that our paper "Fairness in Ranking under Uncertainty" has been accepted to #NeurIPS2021. This is joint work with my advisor Thorsten Joachims and David Kempe from USC. Looking forward to sharing the camera ready version soon! 📄

Emergent behavior papers are always cool but this is also very timely! Econ and AI should be best friends, especially when online platforms continue to scale and evolve riding on the successes and advances in AI.

AI developers building increasingly large pattern recognizers and claiming they are about to solve AGI is like me claiming the ladder I'm building is about to reach the moon.

Such an honor to offer the Baccalaureate Address at Stanford University's Commencement this weekend. I shared what I've learned from Sikh teachings: How a small shift in our perspective can help us see the good in one another—and that enables us to taste the sweetness of life.

The Danger of Minimalist Design (& the death of detail) A short thread...

Recommender systems change your preferences and beliefs by exposing you to content. When trained with RL, these systems have the incentive to manipulate your preferences so they are easier to satisfy. What can we do about this? Our #ICML2022 📑: arxiv.org/abs/2204.11966 1/n

Tech layoffs affect immigrants the most! 5–15% of the 100k+ layoffs are H-1B holders. In 60 days (30-45 in practice), they must find a new job or be deported from the US. Families who've been here 10+yrs are told "We don't need you anymore. Leave." 10+ tips to stay– 1/12

Incredible resource! A much needed course on uncertainty estimation in ML by Aaron Roth

TikTok users feel its algorithm is eerily accurate. But from everything we know about the algorithm, it's a pretty standard recommender system of the kind that every major social platform uses. TikTok's real innovation is something else. Just posted: knightcolumbia.org/blog/tiktoks-s…

Congratulations to Richard Zemel, Ledell Wu, Kevin Swersky, Toni Pitassi, Cynthia Dwork for winning the #ICML2023 Test-of-Time Award on their paper "Learning Fair Representations"! Announcement: icml.cc/Conferences/20… Paper: proceedings.mlr.press/v28/zemel13.ht…

A intriguing thread on the often serendipitous emergent behavior in large ML models! However, it's crucial to recognize that these phenomena can sometimes be negative or inappropriate. The process of diagnosing and successfully addressing them is itself an formidable challenge!

Have you ever done a dense grid search over neural network hyperparameters? Like a *really dense* grid search? It looks like this (!!). Blueish colors correspond to hyperparameters for which training converges, redish colors to hyperparameters for which training diverges.

The competition between LLMs is so intense that you can threaten to leave Claude for ChatGPT, and watch it let go of its guardrails!

Pretty cool. Ask ChatGPT “From all of our interactions what is one thing that you can tell me about myself that I may not know about myself”

📣 📣 📣 Our new paper investigates the question of how many images 🖼️ of a concept are required by a diffusion model 🤖 to imitate it. This question is critical for understanding and mitigating the copyright and privacy infringements of these models! arxiv.org/abs/2410.15002