nsthorat

7 ms·

1.
Lessons of a first time startup founder(nikubaba.com)
2 points by nsthorat 2y ago|0 comments
2.
by nsthorat 3y ago
Lilac co-creator here :) Lilac is an open-source tool that enables AI practitioners to see and quantify their datasets. Lilac allows users to: - Browse datasets with unstructured data. - Enrich unstructured fields with structured metadata u
3.
Lilac: Analyze, structure, and clean unstructured data with AI(lilacml.com)
2 points by nsthorat 3y ago|2 comments
4.
by nsthorat 3y ago
I cant tell if this is research about generative AI, or AI generated research. The new internet sucks
5.
by nsthorat 6y ago
How could a bear population quadruple in a month of lockdown when the gestational period of a bear is 200+ days?
6.
by nsthorat 7y ago
This backend work has already begun (and can run posenet, albeit still slower than WebGL): https://github.com/tensorflow/tfjs/tree/master/tfjs-backend-...
7.
by nsthorat 8y ago
Unfortunately there is no attribution, but this tool was created by Daniel Smilkov, who also built TensorFlow Playground and who is a cocreator of TensorFlow.js. https://twitter.com/dsmilkov
8.
How to build a Teachable Machine with TensorFlow.js, interactive notebook(beta.observablehq.com)
1 points by nsthorat 8y ago|0 comments
9.
by nsthorat 8y ago
There are many reasons to do it in JavaScript: - Many companies and projects have their entire server-side stack in JavaScript and Node.js, and often they want to simply make a prediction through a model. It's quite a lot to ask them t
10.
by nsthorat 9y ago
This probably won't buy you anything. The API you still have is WebGL, and JS is not the bottleneck.
11.
by nsthorat 9y ago
We've done some initial tests ourselves. WASM doesn't yet support SIMD so WebGL tends to be 5-10x faster. SIMD is actively being worked on by many smart people in Chromium / other browsers, so I would expect to see huge wins
12.
by nsthorat 9y ago
Come build a WASM backend for deeplearn.js :)
13.
by nsthorat 9y ago
https://github.com/pair-code/deeplearnjs
14.
by nsthorat 9y ago
Or you improve your algorithms and use the existing hardware (think distributed computing on cheap HDDs).
15.
by nsthorat 9y ago
Why wait?
16.
by nsthorat 9y ago
It doesn't work in node yet, a relevant issue: https://github.com/PAIR-code/deeplearnjs/issues/234
17.
by nsthorat 9y ago
You're right. Some history: We wanted to do hardware accelerated deep learning on the web, but we realized there was no NumPy equivalence. Our linear algebra layer has now matured to a place where we can start building a more functiona
18.
by nsthorat 9y ago
WebGPU conversations are ongoing: https://en.wikipedia.org/wiki/WebGPU WebAssembly is coming along quite nicely. And SwiftShader is a quite nice fallback for blacklisted GPUs. They simulate WebGL on the CPU and take ad
19.
by nsthorat 9y ago
Often times researchers train huge models, but don't think about model size (because they don't have to). We've seen ~200MB production models get down to ~4MB and not lose much precision. I'm quite confident we'll c
20.
by nsthorat 9y ago
There is lots of work being done in model compression (quantization, simple factorization tricks, better conv kernels like depthwise separable convs, etc). We won’t let that happen!
21.
by nsthorat 9y ago
We call ourselves deeplearn.js, but you can use it for general linear algebra! Our NDArrayMath layer is analogous to NumPy, and we support a large subset of it (we support many of the linear algebra kernels, broadcasting, axis reduction, et
22.
by nsthorat 9y ago
This is just the beginning :)
23.
by nsthorat 9y ago
Author of deeplearnjs here. We hear you, and we 100% agree. Stay tuned.
24.
by nsthorat 9y ago
Author of deeplearn.js here. A quick summary: We store NDArrays as floating point WebGLTextures (in rgba channels). Mathematical operations are defined as fragment shaders that operate on WebGLTextures and produce new WebGLTextures. The fra
25.
Cam Arcade: play classic DOS games with your webcam and a neural network(deeplearnjs.org)
2 points by nsthorat 9y ago|0 comments
26.
by nsthorat 9y ago
https://github.com/PAIR-code/deeplearnjs/issues/158
27.
by nsthorat 9y ago
kinda, ya
28.
by nsthorat 9y ago
It works on mobile, it's just slow. Every time we read and write from memory we have to pack and unpack 32 bit floats as 4 bytes without bit shifting operators >.>
29.
by nsthorat 9y ago
Training a neural network on top would require a "proper" training phase, and finding the right hyperparameters that work everywhere turned out to be tricky. Actually, this is what we did originally, in the blog post we'll tr
30.
by nsthorat 9y ago
We're using SqueezeNet ( https://github.com/DeepScale/SqueezeNet ), which is similar to Inception (trained on the same ImageNet dataset) but is much smaller - 5MB instead of inception's 100MB - and inference is

More ›