reading list

need to read more papers ahhh... for research... read more books for life, yikes

(NOTE: "(+)" indicates a would recommend review, since the paper might be useful to others)

##implement_asap

• CACTUs: https://arxiv.org/abs/1810.02334
• MODALS: https://openreview.net/forum?id=XjYgR6gbCEc

foundation_papers

computer_vision

1. SIFT/FAST
2. Alex-Net
3. R-CNN/Fast R-CNN/Mask R-CNN
4. YOLO

NLP

1. Attention is all you need

geometric_DL

1. GCN
2. GAT
3. GraphSAGE

SSL

learning on sets

1. DeepSets
2. PointNet
3. Fabian Fuchs review

generative_modeling

1. Welling VAE
2. Goodfellow GAN

unsupervised_learning

###bias_fairness_in_AI

1. https://dl.acm.org/doi/10.1145/3442188.3445922

projects

#scgat

• scGNN
• Random labels
• feature interactions in NNs, here
• fuzzy jaccard index, FUJI score
• GAT by Velickovic, here
• distinct, Robinson lab on new DGE analysis: https://www.biorxiv.org/content/10.1101/2020.11.24.394213v3

#ekg

• https://www.nature.com/articles/s41591-020-0870-z
• https://www.nature.com/articles/s41598-019-57025-2
• clr survey on ecg: https://arxiv.org/pdf/2103.12676.pdf
• Cactus
• Prototypical networks
• https://www.nature.com/articles/s41591-021-01335-4
• meta pseudolabels: https://openaccess.thecvf.com/content/CVPR2021/papers/Pham_Meta_Pseudo_Labels_CVPR_2021_paper.pdf
• CLOCS (somehow published at ICML!?...) https://icml.cc/Conferences/2021/Schedule?showEvent=8461
• doi:10.4070/kcj.2018.0446; Artificial Intelligence Algorithm for Screening Heart Failure with Reduced Ejection Fraction Using Electrocardiography doi:10.1097/MAT.0000000000001218; Opening the “Black Box” of Artificial Intelligence for Detecting Heart Failure. ASAIO J; Artificial intelligence for the diagnosis of heart failure doi:10.1038/s41746-020-0261-3

#sc2drug

• WGAN review
• similar idea, in Nat Mach Int
• IGMC
• single-cell + drug discovery, a review
• https://www.nature.com/articles/s41467-021-21997-5
• https://www.biorxiv.org/content/10.1101/2021.05.25.445658v2
• https://www.nature.com/articles/s42256-021-00338-7
• https://arxiv.org/abs/2106.02246

#unos/imacs/hcci

• Shapley networks; code here
• organ transplant by vdS; possibly published version: http://proceedings.mlr.press/v139/berrevoets21a.html

#wearables

• classic paper on lstm v. gru
• paper to replicate
• time2vec

safegraph

• https://www.nature.com/articles/s41586-020-2923-3
• https://arxiv.org/abs/2007.03113

#dist_encoder

• MSR for data set similarity
• DeepSets
• set functions for time-series
• applications of data set similarity for bio, here

shared_w/me

• another max welling article on norm flows
• https://arxiv.org/abs/2106.01345
• relevant for DINO, https://arxiv.org/abs/2106.05237
• feature selection, high-dim bio data, https://arxiv.org/abs/2001.08322
• https://arxiv.org/pdf/2106.09643.pdf
• sub-groups: https://arxiv.org/pdf/2103.03399.pdf
• dataset shift: http://proceedings.mlr.press/v130/subbaswamy21a/subbaswamy21a.pdf
• (time-series + causal inference): https://arxiv.org/ftp/arxiv/papers/2107/2107.01353.pdf

fields

XAI

• interpretable GNNs, building on concepts-based explanations: https://arxiv.org/pdf/2107.07493.pdf
• simple way to reduce bias and extract more meaningful information: https://arxiv.org/pdf/1812.10352.pdf

SSL

• CSI
• DINO blog post

tabular_data

• VIME
• TabNet
• SAINT

multi-modal_reps

• https://arxiv.org/pdf/2104.11178.pdf

ML4healthcare

• another vdS, on a tool, https://arxiv.org/abs/2106.04240
• explaining time series, another vdS, https://arxiv.org/abs/2106.05303
• important paper on how AI for covid cxr fails miserably and was not useful, https://www.nature.com/articles/s42256-021-00338-7
• how models with widespread use, such as the EPIC Sepsis Model, actually suck a lot, and we need more competency for ML to be useful: https://jamanetwork.com/journals/jamainternalmedicine/article-abstract/2781307
• https://openreview.net/forum?id=eJIJF3-LoZO
• https://openreview.net/forum?id=h0de3QWtGG
• https://openreview.net/forum?id=unI5ucw_Jk
• https://openreview.net/forum?id=PpshD0AXfA
• alaa branching into biomedical lit, https://www.nature.com/articles/s42256-021-00353-8
• google on predicting gex w/enformer: https://www.biorxiv.org/content/10.1101/2021.04.07.438649v1

ML4biology (includes drug discovery/repurposing, etc.?)

desirable $1$: a simpler classification of fields...

• transcriptome --> small molecule, https://www.nature.com/articles/s41467-019-13807-w
• perturbation modeling, in part with single-cell data: https://doi.org/10.1016/j.cels.2021.05.016
• from Google Accelerated Science team, https://ai.googleblog.com/2020/04/applying-machine-learning-toyeast.html AND https://research.google/pubs/pub49138/ for translational research
• https://www.biorxiv.org/content/10.1101/2021.05.28.446021v1 (Nir Yosef on lineage tracing)
• autoencoder for structural dynamics: https://doi.org/10.1016/j.cell.2015.03.050
• alphafold2: https://www.nature.com/articles/s41586-021-03819-2

time-series

• inception for time-series, https://arxiv.org/pdf/1909.04939.pdf
• sktime, the equivalent of sklearn for time-series, http://learningsys.org/neurips19/assets/papers/sktime_ml_systems_neurips2019.pdf
• voice2series: https://arxiv.org/abs/2106.09296 (ICLR 2021)
• explaining time-series predictions (ICLR 2021), another vdS paper: https://arxiv.org/abs/2106.05303
• causality conditions for time-series by Scholkopf et al.: https://arxiv.org/abs/2005.08543 (ICLR 2021)
• Designing experiments to test pre-training for ts: https://arxiv.org/abs/2110.02095

computer_vision

• great lillian blog post on attention, https://lilianweng.github.io/lil-log/2018/06/24/attention-attention.html
• neural tangent kernel, https://arxiv.org/abs/1806.07572
• random fourier features, https://people.eecs.berkeley.edu/~brecht/papers/07.rah.rec.nips.pdf
• SimSiam, Chen & He, CVPR 2021 (here)
• metadata normalization to remove confounders in metadata (e.g., diff sites), within the network, https://openaccess.thecvf.com/content/CVPR2021/html/Lu_Metadata_Normalization_CVPR_2021_paper.html
• on all the contrastive learning stuff by FAIR: https://openaccess.thecvf.com/content/CVPR2021/html/Feichtenhofer_A_Large-Scale_Study_on_Unsupervised_Spatiotemporal_Representation_Learning_CVPR_2021_paper.html
• SOTA for tumor and organ segmentation as of CVPR 2021, https://openaccess.thecvf.com/content/CVPR2021/html/Zhang_DoDNet_Learning_To_Segment_Multi-Organ_and_Tumors_From_Multiple_Partially_CVPR_2021_paper.html
• better than VQ-VAE, discrete cosine transforms: https://arxiv.org/abs/2103.03841 (ICLR 2021)
• NeRF VAE, geometry of multiple scenes: https://arxiv.org/abs/2104.00587
• limits of pre-training: https://arxiv.org/abs/2110.02095
• Neural Radiant Fields

causal_inference

• resolving latent correlations in disentangled representation nets using weak supervision (another scholkopf): https://arxiv.org/abs/2006.07886
• more comprehensive follow-up than DAGNN (DAG-GNN?) that iclr one, and newer; see: https://arxiv.org/abs/2101.07965
• textbook: https://mitpress.mit.edu/books/elements-causal-inference from Shcolkopf and Jonas Peters (http://web.math.ku.dk/~peters/)
• Melanie Mitchell on abstraction: https://arxiv.org/abs/2102.10717
• Symbolic reasoning in NNs: https://arxiv.org/abs/2102.03406

GRL

• https://arxiv.org/abs/2007.02265 ACM-GCN for improved corr btw node embeddings & topology
• https://arxiv.org/pdf/2106.12575.pdf
• https://arxiv.org/pdf/2103.03212.pdf

AI4good

• https://arxiv.org/abs/2107.10356 reading race in ConvNets on radiological data

###nlp

• https://arxiv.org/abs/2008.09335

###ML4physics

• PDE solvers using GNNs from Welling et al., ICLR'22

Eval metrics

• classic Raschka on it: https://arxiv.org/abs/1811.12808

architectures

transformers

• On Position Embeddings in BERT, see here
• they really are amazing: frozen pretrainted transformer can be trained on lang and do vision tasks w/o fine-tuning: https://arxiv.org/abs/2103.05247
• avoiding degernation of self-attention: https://arxiv.org/abs/2103.03404

contemporary

compsci

ML4bio

• DL for structural bio properties

popsci

• ML for climate change by Bengio
• data privacy
• data privacy again?