JavaScript is not available.

Jun 17, 2022

After 8 months of long coding nights ☕️ we finally officially release Pythae 🥳, a python library unifying generative autoencoder implementations including vaegan🥗, vqvae or RAEs. 🖥️ github repo: github.com/clementchadebe 👉paper: arxiv.org/abs/2206.08309

171

1,000

11h

Large language model #LLM,

@BigscienceW

, #BLOOM 🌸 You've heard these keywords but you don't really know what they mean and where to start? 🫣 With Meriem Bendris we will be holding a webinar on Thursday 2 February at 9am UTC which should give you some answers!

The Pythae library, which brings together many Variational Autoencoder models and enables researchers to make comparisons and conduct reproducible research, is now integrated with Comet ML, allowing...

Comet

@Cometml

Jan 5

Our experiment tracking tool helps you store experiment configs, track training, and compare the results in an easy and understandable way through a visual interface. We're now integrated with the Pythae library for easier ML monitoring!

comet.com

Pythae + Comet

Nov 29, 2022

Meet us at Hall J #1012 booth in 20min to discuss ROOTS 🌸🤗

Quote Tweet

Nov 28, 2022

Wondering how one can create a dataset of several TB of text data to train a language model?

With @BigscienceW, we have been through this exercise and shared everything in our #NeurIPS2022 paper "The BigScience ROOTS Corpus: A 1.6TB Composite Multilingual Dataset"

Show this thread

Nov 28, 2022

Wondering how one can create a dataset of several TB of text data to train a language model?📚 With

@BigscienceW

, we have been through this exercise and shared everything in our #NeurIPS2022 paper "The BigScience ROOTS Corpus: A 1.6TB Composite Multilingual Dataset" 🧵

290

I'm so excited to present my work on VAEs at #NeurIPS2022! 😀 Interested in VAEs? Feel free to stop by our posters: - Thu. 11:30am: A Geometric Perspective on VAEs - arxiv.org/abs/2209.07370 - Thu. 4pm: Pythae: Unifying Generative Autoencoders in Python - github.com/clementchadebe

Pythae - Variational Autoencoder Models with #MLOps interoperability with monitoring tools such wandb, mlflow or comet-ml and model sharing and loading from the HuggingFace Hub Python GitHub: github.com/clementchadebe

Comet

@Cometml

Nov 3, 2022

Pythae is an open source project which makes it easy for users to benchmark different Auto-Encoder models on their custom dataset. Now users can use Comet with Pythae to compare and visualize their models! Shoutout to

and team! Link to the Repo:bit.ly/3WxlZQu

Jean de Nyandwi

@Jeande_d

Oct 30, 2022

Pythae - a great library that provides unified implementations of normal & variational autoencoders(VAE). The library allows inference & training on custom data(with a custom encoder and decoder). Also provides notebooks that walkthrough available models. github.com/clementchadebe

275

Sep 19, 2022

25 models are currently implemented together with 10 sampling methods. It also integrates experiment monitoring tools such as

@weights_biases

and

@MLflow

🧪, allows model sharing and loading from the

@huggingface

Hub🤗 and provides an online documentation and tutorials📚.

I am very happy to share that Pythae, our python library unifying implementations of generative autoencoders, is accepted to #NeurIPS2022 Track on Datasets and Benchmarks🥳 Paper : arxiv.org/abs/2206.08309 Code 🖥️: github.com/clementchadebe

325

Finally, we show that under mild assumptions, the training of the VAE can be kept unchanged and we discuss the link that exists between this metric and the pull back metric. 5/5

Then, we propose to sample from the uniform distribution intrinsically defined on the learned manifold. 4/5

We see the inverse of the covariance matrix in the posterior distribution as a local approximation of the Riemannian metric in the latent space at the embedding point. This allows to build a continuous smooth Riemannian metric on the whole latent space. 3/5

We propose to interpret the latent space of VAEs as a Riemannian manifold and propose to sample uniformly from the learned manifold. We show that it improves the generated samples quality and can benefit more advanced VAE models such as AAE, VAMP-VAE, VAEGAN or IWAE. 2/5

Very happy to announce that our paper "A Geometric Perspective on Variational Autoencoders" just got accepted to #NeurIPS2022 In this paper, we propose a geometrically driven interpretation of Variational Autoencoders. 👉 arxiv.org/abs/2209.07370 1/5

256

I sincerely hope you will enjoy these new features! 😀 This library is also open to new contributors so do not hesitate to reach out if you want to include a model, fix a potential bug 🐛 or would like to see any new feature!

Pythae: Unifying VAE in

@PyTorch

- What's new ? 👉 New models: 25 #opensource implementations now available 👉 Integration of

@MLflow

and

@weights_biases

monitoring tools 👉 Reproduction of original papers 👉 New examples/tutorials (e.g. VAE-LSTM) github.com/clementchadebe

221

merve at GDE summit

@mervenoyann

Aug 28, 2022

some people I look up to (and you might want to follow if you're into ML) 👀

103

MONTREAL.AI

@Montreal_AI

Aug 21, 2022

Pythae: Unifying Generative Autoencoders in Python -- A Benchmarking Use Case Chadebec et al.: arxiv.org/abs/2206.08309 #Artificialintelligence #DeepLearning #MachineLearning

Data Augmentation in High Dimensional Low Sample Size Setting...

Jul 13, 2022

4 - Finally, we validate the method on a medical imaging classification task on ADNI consisting in finding 3D MRIs of Alzheimer disease (AD) patients from cognitively normal (CN) ones. The model again showed useful to augment the data and allowed to improve a SOTA classifier.

3 - Robustness to classifier complexity and dataset size. The model appears able to constantly increase the classification performances of the classifier regardless of it complexity. For the dataset size, the method still allow a performance gain even with more training samples.

Evolution of the accuracy of a benchmark DenseNet classifier according to the number of samples in the train set (i.e. the baseline) (left), the number of parameters of the Densenet (middle).

read image description

ALT

2 - Robustness across classifiers. In addition to a deep neural network (DenseNet), we also tested the method with 1) a Multi Layer Perceptron, 2) a SVM, 3) a kNN and 4) a Random Forest. In each case, the model revealed useful and allowed better classification performances.

Evolution of the accuracy of four benchmark classifiers on reduced balanced MNIST (left) and reduced unbalanced MNIST data sets (right). Stochastic classifiers are trained with five independent runs and we report the mean accuracy and standard deviation on the test set.

read image description

ALT

Then, we conduct a wide experimental study to show the usefulness of the method for Data Augmentation. 1 - Robustness to datasets. Only a fraction of training data is considered (~500) and augmented for each dataset and the classifier (DenseNet) is tested on the original test.

Data augmentation with a DenseNet model as benchmark. Mean accuracy and standard deviation across five independent runs are reported. The first three rows (Aug.) correspond to basic transformations (noise, crop, etc.). The test set is the one proposed in the entire original data set (e.g. ~1000 samples per class for MNIST) so that it provides statistically meaningful results and allows for a good assessment of the model's generalization power.

read image description

ALT

To enhance the generation capability of the model, we also propose a new generation scheme relying on the learned geometry of the latent space. It proves to be very efficient in the context of low sample size setting when compared to the prior or other post-training generations.

VAE sampling comparison. Using either the prior (left) or the proposed generation scheme (right).

read image description

ALT

This model has a the latent space modeled as a Riemannian manifold and combines both Riemannian metric learning and geometry-aware normalizing flows. It better captures the intrinsic geometry of the latent space and leads to convincing interpolations even with small datasets.

Geodesic interpolations under the learned metric in two different latent spaces.

read image description

ALT

Our paper proposing a new #geometry based Variational Autoencoder to perform data augmentation in high dimension low sample size setting just got accepted at IEEE TPAMI! 🥳

@ElinaSutre

@NinonBurgos

👉 Sofware: github.com/clementchadebe 👉Paper: arxiv.org/abs/2105.00026 🧵1/7

arxiv.org

In this paper, we propose a new method to perform data augmentation in a reliable way in the High Dimensional Low Sample Size (HDLSS) setting using a geometry-based variational autoencoder. Our...

Pythae, our library unifying Generative Autoencoder implementations in #Python now integrates the

@huggingface

Hub🤗 so we can share and load models in a few lines of code. 👉Github: github.com/clementchadebe 👉Tutorial: colab.research.google.com/github/clement

111

🌸

BLOOM's intermediate checkpoints have already shown some very cool capabilities! What's great about BLOOM is that you can ask it to generate the rest of a text - and this even if it is not yet fully trained yet! 👶 🧵 A thread with some examples

Quote Tweet

Jun 27, 2022

A milestone soon to be reached

Can't wait to see the capabilities and performance of this long-awaited checkpoint! What about you? Have you already prepared some prompts that you want to test?

twitter.com/BigScienceLLM/…

153

elvis

@omarsar0

Jun 21, 2022

Pythae • cool new python library (by

) that provides implementations of the most common autoencoder models. It provides benchmarking capabilities and allows you to train these models on your own data. github.com/clementchadebe

315

Jun 17, 2022

This library was also used to benchmark 19 generative autoencoders on 5 different tasks (image generation, reconstruction, classification, clustering and interpolation). All the models underwent a hyperparameter search, for 3 dataset and 2 autoencoding architectures.

Pythae currently contains: 🚨21 unified implementations 🎲9 different sampling methods such as a Gaussian mixture model, normalizing flows (MAF, IAF) or PixelCNN 🧩2 pipelines (training and generation) ... making the code easy to use.

Very happy to receive this top contributor award from

@paperswithcode

! 🥳 I am glad to see that my benchmarking library on generative autoencoders has raised some interest in the community. I really hope you will enjoy it 🚀 👉github.com/clementchadebe #DeepLearning #autoencoder

Mar 18, 2022

Just wanted to share with you a small preview of the application we built on the road for 🌸

@BigscienceW

dataset! 🔨 As a #ML practitioner, it was so useful to look at samples and global metrics of our dataset and of the changes made by our processing pipeline! 💡

0:24

9K views

104

Sebastian Raschka

@rasbt

Feb 8, 2022

A treasure trove of various variational autoencoders implemented in PyTorch: github.com/clementchadebe

326

Dec 17, 2021

💡A little video that explains the terms of a supervised training objective of a neural network. I just came across this animation I made a while back when I was learning to use #manim. I thought it was a shame to leave it in a forgotten folder: so it will be a tweet 🤗 #DL #AI

0:33

25.6K views

376

Pierre Glaser

@PierreGlaser

Dec 10, 2021

If you're interested, please come chat with us about our #NeurIPS2021 work on relaxing the KL gradient flow with

@MichaelArbel

and

@ArthurGretton

during today's poster session (8:30am PST)! neurips.cc/virtual/2021/p

GIF

HeKA research team

@TeamHeka

Nov 22, 2021

The first IA & digital healthcare junior seminar will take place Dec. 2, at 4:30 at the

@CRCordeliers

and online, Presentation by Jean Feydy:"Accurate Point Cloud Registration with Robust Optimal Transport". Organisation by

@TeamHeka

junior researchers seminairedoctorantcrc.github.io/aboutme/

Elina Thibeau-Sutre

@ElinaSutre

Nov 22, 2021

Presenting soon ClinicaDL at ANITI Toulouse!

Quote Tweet

ANITI Toulouse

@ANITI_Toulouse

Nov 22, 2021

Nicholas Asher, #ANITI scientific director, opens the 1st edition of the #3iaDWS - a 2 day-workshop with #IA Phd students, post-docs and researchers

Show this thread