[MLDP Newsletter] Machine Learning Communities: highlights and achievements — Feb 2023
Let’s explore highlights and accomplishments of the vast Google Machine Learning communities over the month. We appreciate all the activities and commitment by the community members. Without further ado, here are the key highlights!
Featured Activities by Focus Areas
Keras
Fine-tuning the multilingual T5 model from Huggingface with Keras by ML GDE Radostin Cholakov (Bulgaria) shows a minimalistic approach for training text generation architectures from Huggingface with Tensorflow and Keras as the backend.
Demoing your ML Models using Hugging Face Spaces by ML GDE Sayak Paul (India) showed how to quickly build a DALL-E 2 like system using open-source tools like Diffusers, Gradio, etc.
Focal Modulation: A replacement for Self-Attention by ML GDE Aritra Roy Gosthipaty (India) shares a Keras implementation of the academic paper, Focal Modulation Networks.
How to Use Cosine Decay Learning Rate Scheduler in Keras? by ML GDE Ayush Thakur (India) introduces how you can correctly use the cosine-decay learning rate scheduler using Keras API.
Implementation of DreamBooth using KerasCV and TensorFlow (Keras.io tutorial) by ML GDE Chansung Park (Korea) and Sayak Paul (India) is a project demonstrating DreamBooth technique to fine-tune Stable Diffusion in KerasCV and TensorFlow. Training code, inference notebooks, a Keras.io tutorial, and more are in the repository.
Semantic segmentation with SegFormer and Hugging Face Transformers by ML GDE Sayak Paul (India) discusses how to fine-tune the SegFormer model from Hugging Face Transformers (implemented in TensorFlow) on a custom dataset to generate state-of-the-art results.
Images classification with TensorFlow & Keras (video) by TFUG Abidjan showed how to define an ML model that can classify images according to the category using a CNN. And there was another event, Time series forecasting with TensorFlow (video), sharing how to make time series with TensorFlow.
Hands on Workshop on KerasCV by GDG Hoboken & Stevens Institute of Technology delivered how to use the library to train object detection models using RetinaNet. They also shared about how to generate images with Stable Diffusion, and advanced topics like classifying 3D data and vision transformers.
Hands-on Workshop on KerasNLP by GDG NYC & GDG Hoboken & Stevens Institute of Technology shared how to use pre-trained Transformers (including BERT) to classify text, fine-tune it on custom data, and build a Transformer from scratch.
On-demand ML
Stable diffusion example in an android application — Part 2 by ML GDE George Soloupis (Greece) explains how to get a result from the Diffusion and the Decoder machine learning models and shows the final image at the mobile’s screen (See Part 1 for reference).
Integrate ML in Mobile Application using Flutter 3 Days Flutter Workshop by University of Engineering & Management — Jaipur gave a brief introduction to machine learning and its integration in mobile applications using Flutter.
ML Engineering (MLOps)
Extend your TFX pipeline with TFX-Addons by ML GDE Hannes Hapke (United States) explains how the ML community is building reusable components for TFX.
Semantic Segmentation model within ML pipeline by ML GDE Chansung Park (Korea), Sayak Paul (India), and Merve Noyan (France) was announced as a #TFCommunitySpotlight🏅 winner in February.
[Between the Brackets] AI trends for 2023 by ML GDE Hannes Hapke (United States) and Vikram Tiwari (India) was a tech talk discussing the state of ML in 2023, the anticipated release of OpenAI’s GPT-4 and what this means for the MLOps ecosystem.
Assisting Accountants with Similarity-based Machine Learning by ML GDE Hannes Hapke (United States) explains why ML is crucial for accounting and how to detect categories for banking transactions with similarity-based GoogleDevML models.
Destroy the Dilemma — REST vs gRPC during model serving using TensorFlow Serving by ML GDE Ashmi Banerjee (Germany) walks you through the two popular architectures — REST and gRPC using Python, docker, and TF Serving. She also shared several tutorials related to ML models serving in her Medium.
How to contribute to the MLOps ecosystem through open-source by ML GDE Chansung Park (Korea) explained MLOps and how individuals can contribute to it through open-source. He also shared his best practices from the last two years.
Creating scalable ML solutions to support big techs evolution (slide) by ML GDE Mikaeri Ohana (Brazil) shared how Google can help big techs to generate impact through ML with scalable solutions.
Responsible AI
In The new age of AI: A Convo with Google Brain, ML GDE Vikram Tiwari (India) discussed responsible AI, open-source vs. closed-source, and the future of LLMs.
ML Research
GatedTabTransformer in TensorFlow + TPU / in Flax by Usha Rengaraju was announced as a #TFCommunitySpotlight🏅 winner in February.
Learning JAX in 2023: Part 1 / Part 2 / Livestream video by ML GDE Aritra Roy Gosthipaty (India) covered the power tools of JAX, namely grad, jit, vmap, pmap, and also discussed the nitty-gritty of randomness in JAX.
Image recognition with JAX (slide) by ML GDE Brett Koonce (United States) explained how to use JAX to perform image recognition.
Accelerating Deep Learning research using JAX by Usha Rengaraju (India) introducing features of JAX.
Efficient Task-Oriented Dialogue Systems with Response Selection as an Auxiliary Task by ML GDE Radostin Cholakov (Bulgaria) showcases how, in a task-oriented setting, the T5-small language model can perform on par with existing systems relying on T5-base or even bigger models. And he did a presentation about this paper on ICNLSP 2022.
Solving Unity Environment with Deep Reinforcement Learning by ML GDE Gabriel Cassimiro (Brazil) is a project aiming to train an agent using Deep Q Networks.
UL2 3 new methods for training language models (paper) & Research on Multi-task Cross-Language Model Generation (paper) by ML GDE Qinghua Duan (China) introduces the respective papers.
Paper reviews about DeepMind Chinchilla (paper), Google LaMDA (paper), DeepMind Sparrow (paper) by ML GDE Grigory Sapunov (UK) explain the details of each.
Applied ML
3 Key Ingredients for Embedded Computer Vision Apps by ML GDE Leigh Johnson (United States) covers the key takeaways she learned from building PrintNanny.ai, a monitoring/automation system for 3D printer farms.
Featured Activities by ML Frameworks
#TensorFlow create-tf-app (repository) by ML GDE Radostin Cholakov (Bulgaria) shows how to set up and maintain a machine learning project in Tensorflow with a single script.
#TensorFlow.js Real-time Object Detection in the browser with YOLOv7 and TF.JS (live demo) by ML GDE Hugo Zanini (Brazil) is an implementation of Yolov7 using Tensorflow.js. Hugo open-sourced the code, and it runs directly on the browser.
#TensorFlow Demoing your TensorFlow Models using Hugging Face Spaces by ML GDE Sayak Paul (India) was a workshop discussing how to deploy and demo TensorFlow models with ease using Hugging Face Spaces.
#TensorFlow ChatGPT Prompt Generator by ML GDE Merve Noyan (France). She released a notebook on this app that generates ChatGPT prompts.
#TensorFlow In Machine Learning Meetup hosted by AICamp, ML GDE Nitin Tiwari (India) and Shweta Bhatt (India) participated as a speaker. Nitin shared about end-to-end computer vision models using TF, and Shweta talked about low-resource ML and zero & few shot learning.
#TensorFlow TFUG Kolkata Reboot by TFUG Kolkata was the group’s first event after a long break. This event aimed to make ML/DL with TensorFlow easier to dive into and introduce ML Olympiad 2023.
#TF Hub Building ML Powered Web Applications using TensorFlow Hub & Gradio (slide) by ML GDE Bhavesh Bhatt (India) demonstrated how to use TF Hub & Gradio to create a fully functional ML-powered web application. The presentation was held as part of an event called AI Evolution with TensorFlow, covering the fundamentals of ML & TF, hosted by TFUG Nashik.
#Kaggle The Kaggle Workbook by ML GDE Luca Massaron (Italy) is the newly released book providing examples and exercises to build upon the learning you started with The Kaggle Book.
#Kaggle Enabling Possibilities with Open-Source in Machine Learning (slide) by ML GDE Sayak Paul (India) talked about how open-source is beneficial for an ML practitioner and the community.
#Cloud #BigQuery Thursday Night is Data Night: hands-on Data & ML on Google Cloud by GDG Cloud Belgium was a hands-on session for engineers who have more than a year programming experience. The session covered how to develop and deploy data & ML & AI solutions on the cloud.
#Dialogflow Using ChatGPT to build a chatbot by ML GDE Sachin Kumar (Qatar) is a tutorial explaining how to use ChatGPT to build a chatbot.
News You Can Use by ML GDE Martin Andrews (Singapore) & Building LLM Apps with LangChain by ML GDE (Sam Witteveen) in TFUG Singapore. The speakers in the event focused on techniques for using and training large models and the super-new langchain framework.
MLDP Newsletter — Feb 2023 was originally published in Google Developer Experts on Medium, where people are continuing the conversation by highlighting and responding to this story.
