Alexandru Ginsca

Lead Data Scientist, BearingPoint, France.

DL/ML/CV/NLP.

[Kirk Borne @KirkDBorne]

#Google API can train smaller, faster #AI models: https://t.co/9XizXxW19F by @Kyle_L_Wiggers
——————
#BigData #DataScience #IoT #IoTPL #IIoT #EdgeAI #MachineLearning #DeepLearning #TensorFlow #DataScientists

[Aran Komatsuzaki @arankomatsuzaki]

My first blog post was released 🥳

I have aggregated ~20 notable recent ML papers, esp. from ICLR 2021, with summaries, visualizations and my comments!

The development in each field is summarized, and the future trends are speculated.

https://t.co/E8mGrGHoIB

[Alexandru Ginsca @alexginsca]

@tunguz @kaggle Cool result! What embeddings did you use and how were they normalized?

[AdapterHub @AdapterHub]

We’re happy to present the https://t.co/1lUH2z9YGd framework which lets you train + share adapters with only 2 additional lines of code to your @HuggingFace script https://t.co/KJCpGQNU0u https://t.co/gtfdLz3hl4
@PfeiffJo @arueckle @clifapt @ashkamath20 @seb_ruder @licwu @kchonyc

[Alexandru Ginsca @alexginsca]

Quite a progress!

[AK @_akhaliq]

Are we done with ImageNet?
pdf: https://t.co/WkwBvLSW77
abs: https://t.co/cryNKs6PjT
github: https://t.co/jRKV9aEMRj

[Thomas Wolf @Thom_Wolf]

Surviving every AI wave, two kernels have consistently been the beating hearts of Natural Language Processing:

Datasets and Metrics

Today we release "nlp", a library to easily share & load data/metrics already providing access to 99+ datasets!

Try it👉 https://t.co/37pfogRWIZ

[PyTorch @PyTorch]

FlowTron: Improved Text to Speech Engine from NVIDIA

Try it out now!
Paper: https://t.co/1vMnpQSBbi
Code: https://t.co/O0sYbd2LTX

[Hugging Face @huggingface]

Let’s democratize NLP for all languages! 🌎🌎🌎

Today, with v2.9.1, we are releasing 1,008 machine translation models, covering ` of 140 different languages trained by @jorgtiedemann with @marian, ported by @sam_shleifer. Find your language here: https://t.co/9EMtfopij3 [1/4]

[Hugging Face @huggingface]

Time to push explainable AI 🔬

exBERT, the visual analysis tool to explore learned representations from @MITIBMLab is now integrated on our model pages for BERT, DistilBERT, RoBERTa, XLM & more! Just click on the tag #exbert on @huggingface’s models page: https://t.co/zoJM9okvTW

[Dominik Haitz @d_haitz]

I’ve created a Python package (https://t.co/kacc5hyJLC, 𝘱𝘪𝘱 𝘪𝘯𝘴𝘵𝘢𝘭𝘭 𝘮𝘱𝘭𝘤𝘺𝘣𝘦𝘳𝘱𝘶𝘯𝘬) on top of @matplotlib to conveniently apply the style described here: https://t.co/Y681rp9eNk Feedback & bug reports welcome!

[Suraj Patil @psuraj28]

T5 for summarization is available in @huggingface 's transformers 🔥. And the results are mind-blowing!!!. It's awesome how easy @huggingface makes it to use pre-trained models.

[Andreas Madsen @andreas_madsen]

Using @huggingface Transformer with TextualHeatmap to make an interactive saliency map in Google Colab. Colab link: https://t.co/76uZpyYdjE

[Justin Johnson @jcjohnss]

Today we released code for SynSin, our CVPR'20 oral
that generates novel views from a single image:
https://t.co/3gpHEotZr7

We have:
- Pretrained models
- Jupyter notebook demos
- Training and evaluation
- #pytorch3d integration

Congrats to @oliviawiles1 on the release!

[PyTorch @PyTorch]

Captum is a library for model interpretability. Its algorithms include integrated gradients, conductance, SmoothGrad and VarGrad, and DeepLift. Learn more: https://t.co/IVdye9smGi

[Jeremy Howard @jeremyphoward]

Brilliant job from our amazing community for creating a `timeseries` package for fastai-v2. This package contains many recent best practices and even built-in datasets.
https://t.co/ZmFCmjBdwb

[Kevin Clark @clark_kev]

Excited to share the code and blog post for ELECTRA, a more efficient pre-training method for NLP Transformers. Pre-train a model to good accuracy using only 1 GPU!
Joint work with @lmthang @quocleix @chrmanning
Code: https://t.co/YPQNpEH5m8
Blog Post: https://t.co/6SjC8aKjLH

[Shivon Zilis @shivon]

Fishal recognition

[Ian Soboroff | [email protected] @ian_soboroff]

Google has a pre-cleaned CommonCrawl fork with both news and web subsets. Oh and a new huge model not named after a muppet but after a bacteriophage.

[PyTorch @PyTorch]

Torchmeta is a collection of extensions and data loaders for few-shot learning and meta-learning. It won first place at the Global PyTorch Summer Hackathon last year. Learn more in the blog post from Tristan Deleu, the project author: https://t.co/jFaPaF8GX0

[Ankur Handa @ankurhandos]

PointRend: Image Segmentation as Rendering https://t.co/vVXaIXnF2P

They use adaptive sampling around regions of uncertainty to refine the labels going from coarse to fine resolution inspired by graphics based rendering techniques.

code: https://t.co/R86gZHYksA