Machine Studying Communities: Q3 ‘22 highlights and achievements

Let’s discover highlights and accomplishments of huge Google Machine Studying communities over the third quarter of the yr! We’re enthusiastic and grateful about all of the actions by the worldwide community of ML communities. Listed here are the highlights!

TensorFlow/Keras

Load-testing TensorFlow Serving’s REST Interface by ML GDE Sayak Paul (India) and Chansung Park (Korea) shares the teachings and findings they discovered from conducting load checks for a picture classification mannequin throughout quite a few deployment configurations.

TFUG Taipei hosted occasions (Python + Hugging Face-Translation+ tf.keras.losses, Python + Object detection, Python+Hugging Face-Token Classification+tf.keras.initializers) in September and helped neighborhood members learn to use TF and Hugging face to implement machine studying mannequin to unravel issues.

Neural Machine Translation with Bahdanau’s Consideration Utilizing TensorFlow and Keras and the associated video by ML GDE Aritra Roy Gosthipaty (India) explains the mathematical instinct behind neural machine translation.

Automated Deployment of TensorFlow Fashions with TensorFlow Serving and GitHub Actions by ML GDE Chansung Park (Korea) and Sayak Paul (India) explains easy methods to automate TensorFlow mannequin serving on Kubernetes with TensorFlow Serving and GitHub Motion.

Deploying 🤗 ViT on Kubernetes with TF Serving by ML GDE Sayak Paul (India) and Chansung Park (Korea) reveals easy methods to scale the deployment of a ViT mannequin from 🤗 Transformers utilizing Docker and Kubernetes.

Lengthy-term TensorFlow Steerage on tf.wiki Discussion board by ML GDE Xihan Li (China) supplies TensorFlow steerage by answering the questions from Chinese language builders on the discussion board.

Hindi Character Recognition on Android utilizing TensorFlow Lite by ML GDE Nitin Tiwari (India) shares an end-to-end tutorial on coaching a customized pc imaginative and prescient mannequin to acknowledge Hindi characters. In TFUG Pune occasion, he additionally gave a presentation titled Constructing Pc Imaginative and prescient Mannequin utilizing TensorFlow: Half 1.

Utilizing TFlite Mannequin Maker to Full a Customized Audio Classification App by ML GDE Xiaoxing Wang (China) reveals easy methods to use TFLite Mannequin Maker to construct a customized audio classification mannequin primarily based on YAMNet and easy methods to import and use the YAMNet-based customized fashions in Android initiatives.

SoTA semantic segmentation in TF with 🤗 by ML GDE Sayak Paul (India) and Chansung Park (Korea). The SegFormer mannequin was not accessible on TensorFlow.

Textual content Augmentation in Keras NLP by ML GDE Xiaoquan Kong (China) explains what textual content augmentation is and the way the textual content augmentation characteristic in Keras NLP is designed.

The most important imaginative and prescient mannequin checkpoint (public) in TF (10 Billion params) via 🤗 transformers by ML GDE Sayak Paul (India) and Aritra Roy Gosthipaty (India). The underlying mannequin is RegNet, recognized for its potential to scale.

CryptoGANs open-source repository by ML GDE Dimitre Oliveira (Brazil) reveals easy mannequin implementations following TensorFlow greatest practices that may be prolonged to extra advanced use-cases. It connects the utilization of TensorFlow with different related frameworks, like HuggingFace, Gradio, and Streamlit, constructing an end-to-end answer.

TFX

MLOps for Imaginative and prescient Fashions from 🤗 with TFX by ML GDE Chansung Park (Korea) and Sayak Paul (India) reveals easy methods to construct a machine studying pipeline for a imaginative and prescient mannequin (TensorFlow) from 🤗 Transformers utilizing the TF ecosystem.

First launch of TFX Addons Package deal by ML GDE Hannes Hapke (United States). The bundle has been downloaded a number of thousand instances (supply). Google and different builders preserve it via bi-weekly conferences. Google’s Open Supply Peer Award has acknowledged the work.

TFUG São Paulo hosted TFX T1 | E4 & TFX T1 | E5. And ML GDE Vinicius Caridá (Brazil) shared easy methods to prepare a mannequin in a TFX pipeline. The fifth episode talks about Pusher: publishing your fashions with TFX.

Semantic Segmentation mannequin inside ML pipeline by ML GDE Chansung Park (Korea) and Sayak Paul (India) reveals easy methods to construct a machine studying pipeline for semantic segmentation activity with TFX and numerous GCP merchandise corresponding to Vertex Pipeline, Coaching, and Endpoints.

JAX/Flax

JAX Tutorial by ML GDE Phillip Lippe (Netherlands) is supposed to briefly introduce JAX, together with writing and coaching neural networks with Flax.

TFUG Malaysia hosted Introduction to JAX for Machine Studying (video) and Leong Lai Fong gave a chat. The attendees discovered what JAX is and its basic but distinctive options, which make it environment friendly to make use of when executing deep studying workloads. After that, they began coaching their first JAX-powered deep studying mannequin.

TFUG Taipei hosted Python+ JAX + Picture classification and helped folks study JAX and easy methods to use it in Colab. They shared information concerning the distinction between JAX and Numpy, some great benefits of JAX, and easy methods to use it in Colab.

Introduction to JAX by ML GDE João Araújo (Brazil) shared the fundamentals of JAX in Deep Studying Indaba 2022.

Ought to I alter from NumPy to JAX? by ML GDE Gad Benram (Portugal) compares the efficiency and overview of the problems that will outcome from altering from NumPy to JAX.

Introduction to JAX: environment friendly and reproducible ML framework by ML GDE Seunghyun Lee (Korea) launched JAX/Flax and their key options utilizing sensible examples. He defined the pure perform and PRNG, which make JAX express and reproducible, and XLA and mapping features which make JAX quick and simply parallelized.

Data2Vec Type pre-training in JAX by ML GDE Vasudev Gupta (India) shares a tutorial for demonstrating easy methods to pre-train Data2Vec utilizing the Jax/Flax model of HuggingFace Transformers.

Distributed Machine Studying with JAX by ML GDE David Cardozo (Canada) delivered what makes JAX completely different from TensorFlow.

Picture classification with JAX & Flax by ML GDE Derrick Mwiti (Kenya) explains easy methods to construct convolutional neural networks with JAX/Flax. And he wrote a number of articles about JAX/Flax: What’s JAX?, load datasets in JAX with TensorFlow, Optimizers in JAX and Flax, Flax vs. TensorFlow, and many others..

DDPMs – Half 1 by ML GDE Aakash Nain (India) and cait-tf by ML GDE Sayak Paul (India) have been introduced as Kaggle ML Analysis Highlight Winners.

More energizing on Random Variables, All it’s good to find out about Gaussian distribution, and A deep dive into DDPMs by ML GDE Aakash Nain (India) clarify the basics of diffusion fashions.

In Grandmasters Journey on Kaggle + The Kaggle Ebook, ML GDE Luca Massaron (Italy) defined how Kaggle helps folks within the knowledge science trade and which expertise you need to deal with aside from the core technical expertise.

Cloud AI

How Cohere is accelerating language mannequin coaching with Google Cloud TPUs by ML GDE Joanna Yoo (Canada) explains what Cohere engineers have performed to unravel scaling challenges in giant language fashions (LLMs).

In Utilizing machine studying to rework finance with Google Cloud and Digits, ML GDE Hannes Hapke (United States) chats with Fillipo Mandella, Buyer Engineering Supervisor at Google, about how Digits leverages Google Cloud’s machine studying instruments to empower accountants and enterprise homeowners with near-zero latency.

A tour of Vertex AI by TFUG Chennai for ML, cloud, and DevOps engineers who’re working in MLOps. This session was concerning the introduction of Vertex AI, dealing with datasets and fashions in Vertex AI, deployment & prediction, and MLOps.

TFUG Abidjan hosted two occasions with GDG Cloud Abidjan for college students {and professional} builders who need to put together for a Google Cloud certification: Introduction session to certifications and Q&A, Certification Research Group.

Deploying 🤗 ViT on Vertex AI by ML GDE Sayak Paul (India) and Chansung Park (Korea) reveals easy methods to deploy a ViT B/16 mannequin on Vertex AI. They cowl some important features of a deployment corresponding to auto-scaling, authentication, endpoint consumption, and load-testing.

TFUG Singapore hosted The World of Diffusion – DALL-E 2, IMAGEN & Secure Diffusion. ML GDE Martin Andrews (Singapore) and Sam Witteveen (Singapore) gave talks named “How Diffusion Works” and “Investigating Immediate Engineering on Diffusion Fashions” to carry folks up-to-date with what has been happening on this planet of picture era.

ML GDE Martin Andrews (Singapore) have performed three initiatives: GCP VM with Nvidia set-up and Comfort Scripts, Containers inside a GCP host server, with Nvidia pass-through, Putting in MineRL utilizing Containers – with linked code.

Jupyter Providers on Google Cloud by ML GDE Gad Benram (Portugal) explains the variations between Vertex AI Workbench, Colab, and Deep Studying VMs.

Practice and Deploy Google Cloud’s Two Towers Recommender by ML GDE Rubens de Almeida Zimbres (Brazil) explains easy methods to implement the mannequin and deploy it in Vertex AI.

Analysis & Ecosystem

The primary session of #MLPaperReadingClubs (video) by ML GDE Nathaly Alarcon Torrico (Bolivia) and Girls in Knowledge Science La Paz. Nathaly led the session, and the neighborhood members participated in studying the ML paper “Zero-shot studying via cross-modal switch.”

In #MLPaperReadingClubs (video) by TFUG Lesotho, Arnold Raphael volunteered to steer the primary session “Zero-shot studying via cross-modal switch.”

ML Paper Studying Golf equipment #1: Zero Shot Studying Paper (video) by TFUG Agadir launched a mannequin that may acknowledge objects in photos even when no coaching knowledge is offered for the objects. TFUG Agadir ready this occasion to make folks excited by machine studying analysis and supply them with a broader imaginative and prescient of differentiating good contributions from nice ones.

Opening of the Machine Studying Paper Studying Membership (video) by TFUG Dhaka launched ML Paper Studying Membership and the group’s plan.

EDA on SpaceX Falcon 9 launches dataset (Kaggle) (video) by TFUG Mysuru & TFUG Chandigarh organizer Aashi Dutt (presenter) walked via exploratory knowledge evaluation on SpaceX Falcon 9 launches dataset from Kaggle.

Introduction to MRC-style dialogue summaries primarily based on BERT by ML GDE Qinghua Duan (China) reveals easy methods to apply the MRC paradigm and BERT to unravel the dialogue summarization drawback.

Plant illness classification utilizing Deep studying mannequin by ML GDE Yannick Serge Obam Akou (Cameroon) talked on plant illness classification utilizing deep studying mannequin : an finish to finish Android app (open supply venture) that diagnoses plant ailments.

Nystromformer Github repository by Rishit Dagli supplies TensorFlow/Keras implementation of Nystromformer, a transformer variant that makes use of the Nyström methodology to approximate normal self-attention with O(n) complexity which permits for higher scalability.