1/ Dear graduate students who did not get the U.S. National Science Foundation #NSFGRFP grant: Social medial and your email will be filled with gleeful notes about students who got the award. It hurts to see that you were not chosen. Here are my three rules for dealing with grant failures: #PhDchat #Protip

In our paper Burst Control, we model the L5PC and analyze the conditions of synaptic activation in the apical and basal dendrites for burst generation. We found a novel burst-class, defined inhibitory control and correlated it with limiting plasticity. biorxiv.org/content/10.110…

I've spent a lot of time investigating what goes on inside neural networks. Working on this project was the highest density of mindblowing discoveries since working on DeepDream. Maybe more so.

openai.com/blog/multimoda…🤔

So, where do the interpretable AI kids hang out? I know of SIGKDD 2025 and possibly IJCAIconf, but are there other communities for explainable AI, preferably with emphasis on applications for scientific discovery?

Fun fact! Did you know that when your eyes are tired, CODA_VISIBLE_DEVICES looks just like CUDA_VISIBLE_DEVICES, and you get to spend half an hour decreasing batch sizes ad absurdum until you realize your out-of-memory gpu is actually a typo?

If budgets for NIH and NSF increase dramatically... could we talk in advance about not using the money to train more PhD students and postdocs, and instead figure out how to systematically encourage staff scientist positions? Ideas beyond just individual PI decisions?

Proposal: When teaching causal inference, use the example of kids blowing on birthday candles instead of a firing squad, and ask who gets to make a wish. Same graph, more cake, everyone is happy (spoiler: the parent who gave the go signal gets to make the wish!)

MDETR - Modulated Detection for End-to-End Multi-Modal Understanding pdf: arxiv.org/pdf/2104.12763… abs: arxiv.org/abs/2104.12763 MDETR, an end-to-end modulated detector that detects objects in an image conditioned on a raw text query, like a caption or a question

Just realized "Theory or experiment" is the academic version of "Truth or dare"

I wonder if someone found a correlation between the introduction of playback speed and slower speech rate in video tutorials. I can swear the more time passes, the slower people speak when on regular speed

(1/n) Just finished teaching #CHI2021 course Intro to #ExplainableAI. Definitely a highlight of my CHI experience! Thrilled to have nearly 200 people registered and helped shaping the course Slides: qveraliao.com/xai_tutorial.p… Website: hcixaitutorial.github.io A thread:

PSA: It is OK to ask both your Palestinian AND your Israeli friends and colleagues how they and their families are doing in this horrible time. Showing empathy does not equal taking sides

This is just a reminder that if you have students/colleagues/friends with family in Palestine or Israel, and especially in Gaza, today is a good time to check in with them. Also: they may tell you what's up and that'll make your day worse. But that's not a reason to not do it.

My and Menachem Kalmenson's #GradientClusteron paper is out in PLOS CB! Learn how a model #neuron with a single #dendrite can solve #MachineLearning tasks by making synapses attract and repel with a gradient descent rule! With Idan Segev and ELSC Brain. journals.plos.org/ploscompbiol/a…

Ray Kurzweil wrote in 2014 that a single neuron is more complex than the entire cortex. So how complex is it? see our recent Neuron paper by David Beniaguev and Idan Segev at: authors.elsevier.com/c/1dYq23BtfGzk…

Out work about "Single Cortical Neurons as Deep Artificial Neural Networks" was finally published in Neuron With Mickey London and Idan Segev Neuron paper: bit.ly/3CJic8E code: bit.ly/3CJUeKv data: bit.ly/3xIm6uW All new results in thread below

This may revolutionize data science: we introduce TabPFN, a new tabular data classification method that takes 1 second & yields SOTA performance (better than hyperparameter-optimized gradient boosting in 1h). Current limits: up to 1k data points, 100 features, 10 classes. 🧵1/6

3 years ago my teammates and I set out toward a goal that seemed like science fiction: to build an AI that could strategically outnegotiate humans *in natural language* in Diplomacy. Today, I’m excited to share our Science paper showing we’ve succeeded! 🧵