fastai: A Layered API for Deep Learning

By Jeremy Howard, Sylvain Gugger. Inf., 2020

1,032 citations recorded by the source corpus.

fastai is a deep learning library which provides practitioners with high-level components that can quickly and easily provide state-of-the-art results in standard deep learning domains, and provides researchers with low-level components that can be mixed and matched to build new approaches. It aims to do both things without substantial compromises in ease of use, flexibility, or performance. This is possible thanks to a carefully layered architecture, which expresses common underlying patterns of many deep learning and data processing techniques in terms of decoupled abstractions. These abstractions can be expressed concisely and clearly by leveraging the dynamism of the underlying Python language and the flexibility of the PyTorch library. fastai includes: a new type dispatch system for Python along with a semantic type hierarchy for tensors; a GPU-optimized computer vision library which can be extended in pure Python; an optimizer which refactors out the common functionality of modern optimizers into two basic pieces, allowing optimization algorithms to be implemented in 4–5 lines of code; a novel 2-way callback system that can access any part of the data, model, or optimizer and change it at any point during training; a new data block API; and much more. We used this library to successfully create a complete deep learning course, which we were able to write more quickly than using previous approaches, and the code was more clear. The library is already in wide use in research, industry, and teaching.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition

View Paper: https://www.semanticscholar.org/paper/91e611c3e8705002438fb4439733e47ddec85b5d

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Computer Vision and Pattern Recognition

Citations in the corpus

Avalanche: an End-to-End Library for Continual Learning (2021, background): “plugins, loggers, benchmarks) can also be found in more general machine learning libraries such as PyTorch Lighting [13] and fastai [20].”
SpeechBrain: A General-Purpose Speech Toolkit (2021, methodology): “The Brain.fit() method is inspired by similar methods in libraries such as Scikit-learn [70], Scipy [71], Keras [72], fastai [73], and PyTorch Lightning [74].”
ImageNet-21K Pretraining for the Masses (2021, methodology): “Step 3 - image resizing: ImageNet-1K training usually uses crop-resizing [19] which favours loading the original images at full resolution and resizing them on-the-fly.”
COVID-ResNet: A Deep Learning Framework for Screening of COVID19 from Radiographs (2020, methodology): “The entire data preprocessing, data augmentation and training performed use the fastai [11] framework.”
Erasing Concepts from Diffusion Models (2023, methodology): “To measure the effect of η on interference, we test three different "nudity" erased ESD-u-η models’ performance on 1000 images per each of the 10 Imagenette [19] classes.”
AutoGluon-Tabular: Robust and Accurate AutoML for Structured Data (2020, methodology): “Inspired by Howard & Gugger (2020), a discrete feature with k unique categories observed in the training data is processed by an embedding layer of size: 1.6ˆ k0.”
An Introductory Review of Deep Learning for Prediction Models With Big Data (2020, background): “…(CNN) R & python: Keras (Chollet, 2015) MXNet (Chen et al., 2015), Tensorﬂow (Abadi et al., 2016), h2O (Candel et al., 2015), fastai (python) (Howard and Gugger, 2018) Deep Belief Network (DBN) RcppDL (R) (Kou and Sugomori, 2014), python: Caffee (Jia et al., 2014), Theano (Theano…”
Fast, Accurate, and Simple Models for Tabular Data via Augmented Distillation (2020, background): “This network uses a separate embedding layer for each categorical feature which helps the network separately learn about each of these variables before their representations are blended together by fully-connected layers [59, 63].”
Unified Concept Editing in Diffusion Models (2023, methodology): “2 Erasing Objects To demonstrate the capability of our method to erase objects from the diffusion model’s learned concepts, with potential applications for removing harmful symbols and content, we conducted experiments erasing Imagenette [17] classes, a subset of Imagenet classes [9].”
One Label, One Billion Faces: Usage and Consistency of Racial Categories in Computer Vision (2021, methodology): “To set the learning rate, we use the one-cycle policy [43] using the implementation in [23].”
Blind Backdoors in Deep Learning Models (2020, background): “ai [33], all include multiple lossvalue computations specific to complex image and language tasks.”
SalUn: Empowering Machine Unlearning via Gradient-based Weight Saliency in Both Image Classification and Generation (2023, methodology): “In addition, class-wise forgetting is also considered for SD on the Imagenette dataset (Howard & Gugger, 2020) to prevent image generation from a specified text prompt, which is given by ‘an image of [class name]’.”