Erik Storrs

Segment Anything: Explained

Segment Anything (SAM) proposes a flexible foundation model for image segmentation. It can respond to query points, boxes, masks, or text, and return masks corresponding to those prompts. Curcially, this design allows for zero or few-shot transfer learning to a wide variety of tasks.

Prov-GigaPath: A whole-slide foundation model for digital pathology from real-world data - A Paper Walkthrough

Deep learning with whole-slide images (WSIs) is difficult: each WSI contains tens of thousands of image tiles.

Paper Walkthrough - LongNet: Scaling Transformers to 1,000,000,000 Tokens

One of the biggest bottlenecks in Transformer models is the self-attention mechanism, which has a nasty quadratic computational cost with respect to the number of input tokens. For every token in the input, the model c

Deep Learning

DINOv2: Explained

DINOv2 is a continuation of the original DINO framework published by Meta in 2021. It improves self-supervised visual learning by producing more semantically meaningful patch-level features—without the need for labeled training data.

Deep Learning

Deep learning explainer: a simple single cell classification model

I go through a paired code/text explainer for a basic, from-the-ground-up deep learning model to classify single cell data. The goal here is to cover some basic deep learning theory, with code to go along.

Deep Learning

Enformer: Effective gene expression prediction from sequence by integrating long-range interactions

Paper walkthrough of the Enformer introduced in "Effective gene expression prediction from sequence by integrating long-range interactions" by Avsec et al.

Deep Learning

FixMatch, FlexMatch, and Semi-Supervised Learning (SSL)

In today's post I cover semi-supervised learning (SSL), specifically in the context of the FixMatch and FlexMatch algorithms.

Deep Learning

Paper Walkthrough: P-Net - a biologically informed deep neural network for prostate cancer discovery

In which I cover a new biologically informed neural network architecture.

Deep Learning Featured

Explained: Multi-head Attention (Part 2)

Part 2 of 2 of an explanation of multi-head attention.

Deep Learning Featured

Explained: Multi-head Attention (Part 1)

Part 1 of 2 of a series of posts on attention in transformers. In part 1 I go over the basics of the attention mechanism.

Deep Learning

Explained: Attention Visualization with Attention Rollout

In today's post I cover a simple method for visualization of attention in Transformers.

Deep Learning

Paper Walkthrough: A Neural Algorithm of Artistic Style

Where I cover one of the first deep learning algorithms for neural style transfer.

Deep Learning

Explainer: CNNs and ResNets - Part 2

Part 2 of an explainer on the architecture of a convolutional neural networks and ResNets

Explainer

Explainer: Convolutional Neural Networks (CNNs) - Part 1

An attempt at a visual explanation of convolutions and the basic philosophy behind convolutional neural networks (CNNs).

Deep Learning

Paper Walkthrough: Cross-Covariance Image Transformers (XCiT)

Where I cover Cross-Covariance Image Transformers (XCiT).

Deep Learning

Paper Walkthrough: Bidirectional Encoder Representations from Transformers (BERT)

This week, I cover Bidirectional Encoder Representations from Transformers (BERT).

Bioinformatics Featured

Explained: Deep learning methods for scRNA-seq clustering

In which I cover DESC and scDCC (a revamped version of it's predecessor scDeepCluster); two popular deep learning-based methods for embedding single cell RNA-seq (scRNA-seq) data.

Explainer

Explained: Singular Value Decomposition (SVD)

In which I attempt an intuitive explanation of singular value decomposition (SVD).

Paper Walkthrough

Explained: Principle Component Analysis (PCA)

In this post I'll attempt to give an intuitive, visual explanation of what principle components are and how they are derived.

Deep Learning

Paper Walkthrough: DINO

In this post, I'll cover Emerging Properties in Self-Supervised Vision Transformers by Caron et al., which introduces a new self-supervised training framework for vision transformers.

Deep Learning

Paper Walkthrough: CaiT (Class-Attention in Image Transformers)

In this post, I cover the paper Going deeper with image transformers by Touvron et al, which introduces LayerScale and Class-Attention Layers.

Deep Learning Featured

Paper Walkthrough: DeiT (Data-efficient image Transformer)

Walkthrough of the paper Training data-efficient image transformers and distillation through attention from Touvron et al. that introduces a new distillation for visual transformers.

Deep Learning Featured

Paper Walkthrough: ViT (An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale)

Visual walkthrough of ViT by Dosovitskiy et al. that adapts the Transformer architecture to computer vision, showing that convolutions are not necessary for SOTA performance in vision tasks.

Deep Learning

Paper Walkthrough: Attention Is All You Need

In this post I walk through the classic paper Attention Is All You Need