We also open-sourced our analysis pipeline, collaborating with scholars at the University of Oxford:
— DeepMind (@DeepMindAI) October 17, 2019
https://t.co/ngsj1KVkHV
We also open-sourced our analysis pipeline, collaborating with scholars at the University of Oxford:
— DeepMind (@DeepMindAI) October 17, 2019
https://t.co/ngsj1KVkHV
Code on GitHub: https://t.co/Ru0t4XTH1o
— François Chollet (@fchollet) October 16, 2019
Both #rstats and #python used in the GitHub repo for this important analysis #ddj https://t.co/BkwMKebMcc https://t.co/GEzZ9fnWUX
— Sharon Machlis (@sharon000) October 13, 2019
Mish: Self Regularized Non-Monotonic Activation Function
— ML Review (@ml_review) October 13, 2019
By @DigantaMisra1
𝑓(𝑥)=𝑥⋅𝑡𝑎𝑛ℎ(𝑠𝑜𝑓𝑡𝑝𝑙𝑢𝑠(𝑥))
Increased accuracy over Swish/ReLU
Increased performance over Swish
Githubhttps://t.co/RYzuj0xhDN
ArXivhttps://t.co/YJKTd4yKvr pic.twitter.com/zlyQ0hwggt
We’ve released pre-trained BigBiGAN representation learning models https://t.co/Rhm94rOuX5
— DeepMind (@DeepMindAI) October 8, 2019
on TF Hub: https://t.co/E18skH2iRC
Try them out in a Colab at: https://t.co/ixQZJaABRJ pic.twitter.com/Hu7vPpLkgL
In parallel with this paper, @facebookai has released higher, a library for bypassing limitations to taking higher-order gradients over an optimization process.
— Edward Grefenstette (@egrefen) October 7, 2019
Library: https://t.co/U5dFLBXTHZ
Docs: https://t.co/2mYODGdI8x
Contributions very welcome. https://t.co/F8S7TsZlfe
Due to popular demand, code to reproduce our "Beyond BLEU" paper is now available. Check it out to train your MT models on semantic objectives: https://t.co/C0crMwd25F https://t.co/Exz5pQcUSk
— Graham Neubig (@gneubig) October 4, 2019
@RSNA Intracranial Hemorrhage Detection @kaggle competition starter pack using @fastdotai
— Radek Osmulski (@radekosmulski) September 20, 2019
✅github repository: https://t.co/BLlNEuKlmk
✅competition website: https://t.co/HcP3HTUPgD
✅approach overview: https://t.co/0aNSrxFyRn
✅@fastdotai forum post: https://t.co/6Cag5b5v1m pic.twitter.com/hFmE0mG1Q0
“Espresso achieves state-of-the-art ASR performance on the WSJ, LibriSpeech, and Switchboard data sets among other end-to-end systems without data augmentation, and is 4--11x faster for decoding than similar systems (e.g. ESPnet).” #SpeechTech https://t.co/ZvvuwOGtTT
— Delip Rao (@deliprao) September 20, 2019
Shape and Time Distortion Loss for Training Deep Time Series Forecasting Models
— Thomas Lahore (@evolvingstuff) September 20, 2019
paper: https://t.co/jgRT3tC5Df
code: https://t.co/NIC6XjXlyn pic.twitter.com/dopIL5qYYu
Some more🔥for y'all:
— Tim Vieira (@xtimv) September 16, 2019
🔥Now 10x speedup over numpy w/ the "Exp-sort trick" (the cooler cousin of the "Gumbel-max trick")
🔥Exp-sort trick is trivial to parallelize/vectorized
🔥Give a Gumbel-sort trick for log weights
🔥1-pass algs that work w/streamshttps://t.co/su0ufYS67h https://t.co/X39gGfa48w
We (@iansimon + @huangcza + @jesseengel + @fjord41 + @notwaldorf) have now released a Colab notebook where you can play around with some Transformer models for generating piano performances: https://t.co/AQnV7jsYxM
— Ian Simon (@iansimon) September 16, 2019
And the models are trained on transcribed YouTube recordings!