ImageNet, every mention
24 scenes, the whole family · ← back to ImageNet
tap a year for its mentions
every year anyone Ari Morcos 6Michael Royzen 3Jeff Dean 3Justin Johnson 2Jeremy Howard 2Fei-Fei Li 2Shawn Wang 1RJ Honicky 1Rishabh Agarwal 1Peter Robicheaux 1
Verbatim, from the transcripts: the passages where ImageNet comes up
Simulating Humanity: from Generative Agents to 8 Billion Digital Twins — Joon Sung Park, Simile AI
- ▶ 1:08:14 Joon Sung Park So, of course, Michael was one of the co-authors of the ImageNet, kicks her the AI revolution back in 2013, has been instrumental in human-centered AI.
🔬 Training Transformers to solve 95% failure rate of Cancer Trials — Ron Alfa & Daniel Bear, Noetik
- ▶ 12:46 unnamed speaker And I mean, like a good comparison is to the ImageNet data set, which kicked off 2 times in the scene
The AI Frontier: from Gemini 3 Deep Think distilling to Flash — Jeff Dean
- ▶ 3:52 Jeff Dean Yeah, I mean, I think distillation was originally motivated because we were seeing that we had a very large image data set at the time, you know, three hundred million images that we could train on with, you know, I forget, like 20,000…
- ▶ 57:58 Jeff Dean Yeah, I mean, I, I did some work on a early model that fused together a language-based model where you have, you know, nice word-based representations, and then an image model where you have trained it on image net-like things.
- ▶ 1:06:17 Jeff Dean It gave us a 70% relative error improvement in image net 22 K, which is the 22,000 category thing.
SAM 3: The Eyes for AI — Nikhila & Pengchuan (Meta Superintelligence), ft. Joseph Nelson (Roboflow)
- ▶ 1:01:11 unnamed speaker We were interviewing a bunch of robotics folks here as well as like, uh, Fei-Fei Li, who obviously started ImageNet.
After LLMs: Spatial Intelligence and World Models — Fei-Fei Li & Justin Johnson, World Labs
- ▶ 2:10 Justin Johnson Watching all the ImageNet excitement around AlexNet at that, that quarter. 2 times in the scene
- ▶ 3:32 unnamed speaker So if you think about AlexNet, the core pieces of it were obviously ImageNet. 3 times in the scene
- ▶ 14:05 Fei-Fei Li He and I were looking at what is beyond ImageNet object recognition, and at that time, we, you know, the convolutional neural network was, uh, has proven some power in ImageNet tasks, so, so ConvNet is a great way to represent images. 2 times in the scene
Why RL Won — Kyle Corbitt, OpenPipe (acq. CoreWeave)
- ▶ 54:21 Shawn Wang Because it is like removes more and more of the human judgment and like the history of machine learning all the way from like, I guess the, the, the, the, the start of like, uh, uh, image net and everything, uh, is, is really like that,…
Better Data is All You Need — Ari Morcos, Datology
- ▶ 11:37 Ari Morcos So if you think about research circa, say, 2018, given ImageNet maximized performance on the Val set or on the test set, right? 3 times in the scene
- ▶ 13:24 Ari Morcos That meant that we went from image net, a million data points to literally trillions of tokens, a million fold increase in data quantity in a matter of like several years.
- ▶ 51:14 Ari Morcos So there wasn't a clear incentive to actually go and do these hard experiments to try to figure out how to make a good curriculum, because like, who cares if I can get you to image net performance in 80 epochs instead of a 160 epochs?
- ▶ 1:05:58 Ari Morcos There was like an iClear best paper from 20 17 that showed this, that people were really surprised that, that models could memorize all of ImageNet.
Information Theory for Language Models: Jack Morris
- ▶ 1:09:38 Jack Morris Deep Neural Networks with AlexNet, which I think was like, 20, 10 to 20 12 era, where we just started training on ImageNet, which is like a scale no one had ever seen before of millions of images.
The Magic of LLM Distillation — Rishabh Agarwal, Google DeepMind
- ▶ 1:21 Rishabh Agarwal That has happened since then, and especially now we're talking about LLMs back then, we were talking about classifiers, like, which will, it's a train on ImageNet and what say, so I thought it would be a good idea to give a tutorial, at…
Best of 2024 in Vision [LS Live @ NeurIPS]
- ▶ 37:21 Peter Robicheaux This is the ImageNet classification accuracy, but yeah, it, it does better if you increase the resolution, which means that it's actually leveraging and finding, um, fine-grained visual features.
[Paper Club] Intro to Diffusion Models and OpenAI sCM: Simple, Stable, Scalable Consistency Models
- ▶ 45:47 RJ Honicky And then, um, and then, you know, for a conditional class, uh, image net, 64 by 64, it's a different one, but it's also in this joint training.
In the Arena: How LMSys changed LLM Benchmarking Forever
- ▶ 30:25 Anastasios Angelopoulos The reason why you and others in the community have that, have that instinct is because when you look at something like a benchmark, like an image net, a static benchmark, what happens is that if I give you a million different models that…
llm.c's Origin and the Future of LLM Compilers - Andrej Karpathy at CUDA MODE
- ▶ 0:29 unnamed speaker A founding member of OpenAI, the reference human for ImageNet.
Breaking down the OG GPT Paper by Alec Radford
- ▶ 18:39 unnamed speaker So, for example, you can use ResNet that was trained to classify ImageNet. 2 times in the scene
Beating GPT-4 with Open Source Models - with Michael Royzen of Phind
- ▶ 2:12 Michael Royzen Um, I took, um, this data set called ImageNet. 3 times in the scene
The End of Finetuning — with Jeremy Howard of Fast.ai
- ▶ 16:41 Jeremy Howard I'm sure ImageNet, you know, is going to be an NLP thing as well.
- ▶ 33:48 Jeremy Howard And, uh, specifically it was who can train ImageNet the fastest.