ImageNet
44
MENTIONS
26
EPISODES
26
PODCASTS
Search complete. 44 mentions across 26 episodes found for "ImageNet".
Sep 23, 2026
Gaute EinevollHOST
Mm
Mihai PetroviciGUEST
... like ImageNet, for example, which are naturalistic-
Gaute EinevollHOST
Mm
Mihai PetroviciGUEST
... images, et cetera.
Frank HutterGUEST
Um, like any image is the same, um, in terms of, you know, like there, there's pixels and, uh, you typically have, um, very, very similar, um, spatial re-relationships between the pixels and so on.
Frank HutterGUEST
And, and you can take images-- Like you can have an ImageNet that is just all-- ba-basically, um, cover all kinds of different, um, images in the world and learn on that one dataset.
Frank HutterGUEST
But what you need for tabular data is actually, yeah, a whole lot of different, um, tables because if you have one table from medicine and then you have another table from insurance, there's just n-nothing you can learn in terms of the, um, the individual rows.
Frank HutterGUEST
Um, from one can't tell you anything about the other one.
5 MINS LATER
Frank HutterGUEST
It's not a, a dataset that you can actually reasonably learn a statistical learning algorithm from.
Frank HutterGUEST
And those types of datasets, um, just haven't been there, and we needed to, um, generate them in order to then learn on them.
Tim ScarfeHOST
And just to hammer that home, there has been an ImageNet moment for tabular data, right? So until the day before yesterday, figuratively speaking, deep learning did not work for tabular data, and now it works dramatically better than CatBoost and XGBoost.
Frank HutterGUEST
TaPFN is the first algorithm that's actually been learned from data to be better at what it's supposed to do.
Fei-Fei LiGUEST
Before ChatGPT was released, we formed a center called Language Model, uh, Research Center to put, to put out those benchmarks.
Fei-Fei LiGUEST
You know, you, you would say ImageNet 15 years ago was one of the very first-
Ed LudlowHOST
That's right
Fei-Fei LiGUEST
... benchmarks of AI.
Fei-Fei LiGUEST
Before ChatGPT was released, we formed a center called Language Model Research Center to put out those benchmarks.
Fei-Fei LiGUEST
You know, you would say ImageNet... 15 years ago was one of the very first benchmark of AI.
Fei-Fei LiGUEST
I continue to believe importance of benchmark from independent bodies and public sector bodies like academia, as well as the shared responsibility of industry, of government together.
Ed LudlowHOST
Dr. Fei-Fei Li, the founder and CEO of World Labs, but also active researcher, academic, pioneer in the field of AI.
Harriet FarlowGUEST
We have to tell it exactly what it is based on the data set that we're essentially matching it to.
Harriet FarlowGUEST
So ImageNet is a very big data science library.
Harriet FarlowGUEST
that has thousands and thousands of images.
Harriet FarlowGUEST
And it's basically the way that machine models are trained a lot of the time, at least public ones a few years ago.
Alex and TylerHOST
So when does AI actually enter the picture with NVIDIA? The next really important moment happens in 2012.
Alex and TylerHOST
There was a neural network called AlexNet, and AlexNet performed extremely well in this major image recognition competition called ImageNet.
Alex and TylerHOST
The researchers used NVIDIA GPUs to train it, and this helped to demonstrate just how useful GPUs could be for deep learning.
Alex and TylerHOST
So why are GPUs so good for AI? It's a similar reason they're good for graphics, right? Training an AI model requires an insane amount of math.
David VivancosGUEST
We have seen it in the AI revolution, having in the field for decades, as we speak earlier.
David VivancosGUEST
But when you start working, when you start seeing that everything we have these starting capabilities with ImageNet, for example.
David VivancosGUEST
And overnight, basically, you have the same capabilities that hundreds of thousands of people have done, for example, with language.
David VivancosGUEST
These Microsoft people working on the embedding of languages.
Srinivas NarayananGUEST
And that, you know, this was back in 2012 and we were trying to figure out how to improve our face recognition systems.
Srinivas NarayananGUEST
And we realized that there was something big happening in computer vision around that time when ImageNet happened.
Srinivas NarayananGUEST
So I got exposed to that and, you know, and yeah.
Srinivas NarayananGUEST
For a few years, I sort of still worked on other things.
Gabe PereyraGUEST
I think one strong intuition is when I was doing research, you can publish a research paper on a 1% improvement on a benchmark.
Gabe PereyraGUEST
And a lot of the ImageNet progress came from people just being like, if I initialize a model in this way, and I get this 2% improvement, like that's actually meaningful because I stack all these up.
Gabe PereyraGUEST
But for a user, they can't tell like a 1% difference.
Gabe PereyraGUEST
And even now between like some of these model changes, it's super hard to tell.
Christian CataliniGUEST
We concluded that, look, anything that's measurable will be automated.
Christian CataliniGUEST
We've seen it with ImageNet when Fei-Fei Li started this whole revolution with the image collection.
Christian CataliniGUEST
We see it with self-driving.
Christian CataliniGUEST
Given enough miles, it drives way better than a human.
16 more episodes mention ImageNet.
Create an account to see the whole feed, search across every transcript, and follow the entities you care about.