Going to tweet out a few ICLR submissions that look interesting based on their abstracts: https://openreview.net/forum?id=8mVSD0ETOXl
Here's a neat theoretical paper on the capability of graph networks on the graph coloring problem https://openreview.net/forum?id=uv_x9uAH1MY
This interesting work uses BERT models pretrained on proteomics data to boost performance on downstream tasks https://openreview.net/forum?id=hga6dk7nxFB
This paper uses graph neural networks to learn continuous time PDEs without restrictions on grid choice. Seems potentially useful! https://openreview.net/forum?id=aUX5Plaq7Oy
This paper uses BERT style models to predict protein-drug interactions. Need to come back and read carefully but potentially very interesting https://openreview.net/forum?id=Zqf6RGp5lqf
Ensembling for improving predictive performance on low data problems. Not sure how much I believe it from the abstract but possibly very neat and worth reading https://openreview.net/forum?id=_77KiX2VIEg
Playing Atari with capsule networks with 92% fewer parameters than CNNS. I'm really interested to see more capsule network research mature https://openreview.net/forum?id=GeOIKynj_V
https://openreview.net/forum?id=GeOIKynj_V (forgot link)
Using transformers to learn representations of logs. Log handling with deep learning is a surprisingly hard problem in systems software https://openreview.net/forum?id=-5VpoDCExrU
Using reinforcement learning to find longer proofs in theorem proving https://openreview.net/forum?id=gZ2qq0oPvJR
Of societal interest especially for West coast folks. Wild fire prediction with convnets https://openreview.net/forum?id=aia4HejvBmY
Potentially interesting work on estimation error bounds for metalearning. Need to read more carefully https://openreview.net/forum?id=SZ3wtsXfzQR
A neat paper drawing a link between transformers for protein contact prediction and older Potts model approaches https://openreview.net/forum?id=oVz-YWdiMjt
Improved training procedures for graph convolutional networks on large graphs https://openreview.net/forum?id=Oq79NOiZB1H
Metagenome2vec, an architecture for characterizing metagenomic states with applications in disease characterization. I'd love to see this one get into @deep_chem https://openreview.net/forum?id=IT2s2Ub6skl
Chemistry Question Answering with BERT models. Neat! https://openreview.net/forum?id=oeHTRAehiFF
Counterfactual variational autoencoders. Worth a closer look https://openreview.net/forum?id=D3TNqCspFpM
RNA alternative splicing prediction. With a new dataset and model! https://openreview.net/forum?id=BL4FZG2bCR7
Causal prediction in minkowski space time. Neat intermingling of physics and deep architectures https://openreview.net/forum?id=a7E9TUTXWAT
Improving neural machine translation for African languages https://openreview.net/forum?id=Q5ZxoD2LqcI
Using spectral analysis to understand learning power of graph convolutions https://openreview.net/forum?id=-qh0M9XWxnv
A pipeline for medical time series. Slightly more applied than the usual ICLR work but seems useful https://openreview.net/forum?id=xnC8YwKUE3k
Using AlphaGo Zero to solve NP hard graph problems with graph networks https://openreview.net/forum?id=0_ao8yS2eBw
Using hamiltonian dynamics within neural networks. Neat stuff! https://openreview.net/forum?id=4T489T4yav
Learning mesh based simulation with graph networks. GNNs and physics have some really rich connections https://openreview.net/forum?id=roNqYL0_XP
Copy number variant detection with deep networks. Another one I'd love to see in @deep_chem ! https://openreview.net/forum?id=24-DxeAe2af
Improved training of quantum neural networks. That is a neural network trained on an actual quantum computer! This one looks neat https://openreview.net/forum?id=meG3o0ttiAD
A strategy for improving reference identification in low data neural machine translation https://openreview.net/forum?id=IazZhsJK7wJ
A mass conserving LSTM. Another really neat model baking in physical priors into deep architectures https://openreview.net/forum?id=Rld-9OxQ6HU
MARS, a technique for multi-objective drug discovery! This looks really cool and would love to see in @deep_chem https://openreview.net/forum?id=kHSu4ebxFXY
Using the symplectic structure of Hamiltonian dynamics in deep networks. Another neat physical deep learning paper https://openreview.net/forum?id=B5VvQrI49Pa
The neural vortex method. Another physics inspired deep learning paper https://openreview.net/forum?id=_8EQ_gMAHFy
Neural architecture search for graph networks https://openreview.net/forum?id=IjIzIOkK2D6
Using energy based models to improve state of the art in retrosynthesis https://openreview.net/forum?id=0Hj3tFCSjUd
Using a protein geometric autoencoder to improve modelling of protein dynamics https://openreview.net/forum?id=LxhlyKH6VP
Lagrangian fluid dynamics with graph networks https://openreview.net/forum?id=7WwYBADS3E_
Neural representation and generation for RNA secondary structure https://openreview.net/forum?id=snOgiCYZgJ7
Ok done for now. There's a very large number of papers! There's still a large number I haven't looked at