How does an infinitely wide net perform?
Blog post : Ultra-Wide Deep Nets and Neural Tangent Kernel (NTK)
and the paper : On Exact Computation with an Infinitely Wide Neural Net
1 Like
Transport-Based Neural Style Transfer for Smoke Simulations
Abs : Transfer features from natural images to smoke simulations, enabling general content-aware manipulations ranging from simple patterns to intricate motifs. The proposed algorithm is physically inspired, since it computes the density transport from a source input smoke to a desired target configuration. Temporal consistency is ensured by transporting and aligning subsequent stylized velocities, and 3D reconstructions are computed by seamlessly merging stylizations from different camera viewpoints.
2 Likes
Just saw this paper to help with Data Augmentation for NLP.
Really great data augmentation system using label-conditioned language model generation to supplement data. And using a pretrained classifier to filter generated sentences to be close to the distribution of the original dataset.
5 Likes
Interesting… They’re quoting Jeremy too
WHAT DOES THE AI COMMUNITY THINK?
- “It’s been a long time since we’ve seen a new optimizer reliably beat the old favorites; this looks like a very encouraging approach!” – Jeremy Howard, a founding researcher at fast.ai.
This is an interesting publication by S. Merity about a single headed attention LSTM with promising results for such a small model:
„Single Headed Attention RNN: Stop Thinking With Your Head“
4 Likes
74 Summaries of Machine Learning and NLP Research:
“http://www.marekrei.com/blog/74-summaries-of-machine-learning-and-nlp-research/“
3 Likes
Stumbled upon this, short summaries on selected machine learning papers:
1 Like
URL is broken, gives a 404 error.
The last " is being included in the url
Thanks @urmas.pitsi , this looks like an excellent source. Could you please edit your post to fix the link?
1 Like
Sorry, my bad! Thanks for noticing! I fixed the link hopefully 
2 Likes
Pretty neat blogpost about compressing Bert. Various related papers and results.
1 Like
This is very interesting work on transfer learning for computer vision with medical image data (covers NN architectures and initializations):
2 Likes
Your “deep double descent” reading list:
(Optional but very nice) Intro blog post:
Deep double descendant blog post:
(Optional but very nice) Additional summary/comment:
(You can find the links to the official papers in the blog posts.)
4 Likes
1 Like
A new saliency map algorithm for interpretability:
4 Likes
StyleGANV2 is out. Big changes:
No more progressive growing
New normalization and regularization
3 Likes