[MLDP Newsletter] May2023 — Machine Learning Communities: highlights and achievements

[MLDP Newsletter] May 2023 — Machine Learning Communities: highlights and achievements

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

Implement adaptive test as an auto-encoder by ML GDE Mitsuhisa Ohta (Japan) explains the mechanism of the test which issues appropriate questions to each examinee using AutoEncoder and implementing it.

[ML Story] My Keras Chronicles by ML GDE Aritra Roy Gosthipaty (India) summarized his story of getting into deep learning with Keras. He also jotted down pointers as to how one could get into the open source community.

https://medium.com/media/48f9c330ead1732c20fd0c43cd81ac38/href

DCGANs: Unleashing creativity with TensorFlow by Nitin Tiwari (India) was an engaging workshop that explored the core concepts of DCGANs and practical application through Keras & TensorFlow. Participants gained insights on harnessing the power of Keras to construct compelling generative models.

The Ideal ML Workflow — Doing ML the right way by Ayush Thakur (India) was a talk about the ideal ML workflow and why one should do ML the right way. He touched upon using Keras as a choice of ML framework and W&B (sync TensorBoard) as a choice of MLOps platform. Plus, his article Keras Dense Layer: How to Use It Correctly) explored what the dense layer in Keras is and how it works in practice.

TensorFlow at Google I/O 23: A Preview of the New Features and Tools by TFUG Ibadan explored the preview of the latest features and tools in TensorFlow. They covered a wide range of topics including Dtensor, KerasCV & KerasNLP, TF quantization API, and JAX2TF.

On-device ML

MediaPipe with a bit of Bard by ML GDE Martin Andrews (Singapore) described how MediaPipe fits into the ecosystem, and showed 4 different demonstrations of MediaPipe functionality: audio classification, facial landmarks, interactive segmentation, and text classification.

https://medium.com/media/144a19c68efcd78327402d5f1d874861/href

Implementing ML in our mobile apps with MediaPipe by ML GDE Juan Guillermo Gomez Torres (Mexico) explains how to implement ML models to mobile apps with Google ML Kit and MediaPipe.

Responsible AI

Responsible AI: Principles and Practices by ML GDE Guan Wang (Singapore) introduced Google’s principles and tools on responsible AI and Monetary Authority of Singapore (MAS)’s principles, which is specifically targeted for financial institutes. He also shared efforts to achieve responsible use of AI.

Building a More Inclusive Web — How to create Fair Recommendation Algorithms by ML GDE Ashmi Banerjee (Germany) explored the importance of fairness in recommendation systems and discussed strategies for designing and implementing systems that promote equity.

ML Engineering (MLOps)

Apigee API Management & MLOps with VertexAI by GDG Cloud Belgium shared how to expose your APIs with Google Apigee API Management and how to use MLOps to implement robust ML systems with Vertex AI.

Developing and Deploying Your First Machine Learning Model by TFUG Bhubaneswar and GDG Cloud Bhubaneswar aimed to equip participants with the knowledge and skills to train and deploy machine learning models on Google Cloud. They shared how to choose appropriate models, preprocess data, train effectively, and evaluate performance. They also discovered how to deploy models as APIs for seamless integration into applications and real-time predictions.

ML Research

20 steps to train a deployed version of the GPT model on TPU by ML GDE Jerry Wu (Taiwan) shared how to use JAX and Google Cloud TPU to train and infer Chinese question-answering data.

In May 2023 Meetup hosed by TFUG Mumbai, ML GDE Aritra Roy Gosthipaty (India) and ML GDE Ritwik Raha (India) gave a talk, Decoding End to End Object Detection with Transformers and covered the architecture of the mode and the various components that led to DETR’s inception.

A collaborative deep multitask learning network for face image compliance to ISO/IEC 19794–5 standard by ML GDE Arnaldo Gualberto (Brazil) is a paper in which he has been working with multitask learning for computer vision using TensorFlow. He extended undercomplete autoencoders to employ a multi-and-collaborative learning approach, where supervised and unsupervised learning are performed concurrently and collaboratively.

In Discussion on Google Deepmind & Princeton’s ‘Tree of Thoughts’ method by GDG Cloud Silicon Valley, GDG Fremont, GDG Cloud Fremont, GDG Houston, and GDG Austin, people explored Tree of Thoughts (ToT) approach (paper) and its strategic decision-making capabilities and improved problem-solving prowess.

LLM-related activities

#PaLM2 Introduction to Google’s PaLM 2 API by ML GDE Hannes Hapke (United States) introduced how to use PaLM2 and summarized some major advantages of it.

#PaLM2 Multimodal Transformers — Custom LLMs, ViTs & BLIPs by TFUG Singapore looked at what models, systems and techniques have come out recently related to multimodal tasks, with talks on: 1) Images+LLMs : Advances in Multimodal Models, 2) Transformers Agents and Building your own Multi Modal system, and 3) Using LLMs to perform Job Grading and Evaluations. In particular, ML GDE Sam Witteveen (Singapore) looked into various multimodal models and systems and how you can build your own with the PaLM2 Model.

#PaLM2 Paper review: PaLM 2 Technical Report by ML GDE Grigory Sapunov (UK) looked into the details of PaLM2 and the paper.

#PaLM2 In Attention Mechanisms and Transformers by GDG Cloud Saudi, ML GDE Ruqiya Bin Safi (Saudi Arabia) talked about Attention and Transformer in the field of natural language processing. The group also hosted Hands-on with the PaLM2 API to create smart apps(Jeddah) and explored what LLMs, PaLM2, and Bard are; how to get and use PaLM2 API; and how to create smart apps using PaLM2 API.

#Bard Bard for Students by Istanbul Kultur University helped students understand how to use the Bard tool effectively.

#Bard STI CDO meets Google Bard by STI College Cagayan de Oro introduced Bard and how it can be used to solve real-world problems. They showed a demo using Bard to see how it can generate text, translate languages, and write different kinds of creative content.

#Dialogflow Building Bots for the Real World by ML GDE Yüksel Tolun (Turkey) presented how Dialogflow CX could help developers build bots that are marketable and reliable in a world where LLMs are becoming more popular. He shared technical implementation details and best practices.

Unraveling the mystery of the transformers by ML GDE Moises Martinez (Spain) described what a Transformer is, how it works, how we can build or reuse one to build NLP or computer vision systems and see how they fit into the architectures of the new cognitive systems that seem to be revolutionizing the world of AI.

TFUG Chennai May 2023 Meetup by TFUG Chennai shared knowledge on NLP, LLM and AI governance. Demystifying LLM with introduction about the fundamentals; Harnessing the power of NLP and real-time use cases in action (using TF-IDF model); and AI Security Governance were main topics.

In Generative AI on Google Cloud followed by Q & A by GDG Cloud Indy, Googler Chris Wetnight participated as a speaker and shared the latest technologies in Gen AI.

Crash Course: Building Large Language Model at Scale by ML GDE Jeongkyu Shin (Korea) was a 6-hour crash course covering the entire process of building massive language models. He explained the underlying transformer-based model algorithms, distributed processing technology, GPUs and NPUs, and TPUs along with the rapid evolution of language models. As a codelab, he trained GPT-2 and T5 based models with 2B~7B parameters on actual multi-node servers.

Featured Activities by ML Frameworks

#TensorFlow OSINT Investigation on LinkedIn by ML GDE Mathis Hammel (France) investigated a network of fake companies on LinkedIn with a total reach of 250k+ people. He leveraged a homemade tool based on a TensorFlow model and hosted it on Google Cloud. Some technical explanations of generative neural networks were also included. More than 690K viewed his Twitter thread and it got 1200+ RT and 3100+ Likes.

#TensorFlow The Lord of the Words : The Return of the experiments with DVC (slides) by ML GDE Gema Parreno Piqueras (Spain) was a talk to explain Transformers in the neural machine learning scenario, and how to use Tensorflow and DVC. In the project, she used Tensorflow Datasets translation catalog to load data from various languages, and TensorFlow Transformers library to train several models.

#TensorFlow Accelerate your TensorFlow models with XLA and Ship faster TensorFlow models with XLA by ML GDE Sayak Paul (India) shared how to accelerate TensorFlow models with XLA (slides) in Cloud Community Days Kolkata 2023 and Cloud Community Days Pune 2023.

#TensorFlow Building ML Powered Web Applications using TensorFlow Hub & Gradio by ML GDE Bhavesh Bhatt (India) demonstrated how one can use TensorFlow Hub & Gradio to create a fully functional ML powered web application in Cloud Community Day Indore 2023.

#TensorFlow TensorFlow for Computer Vision: Applications and Techniques by ML GDE Nitin Tiwari (India) delved into how TensorFlow can be used for computer vision tasks like image classification, object detection, and segmentation. He helped the attendees understand the working of CNN and some real-world applications of CV in various industries such as healthcare, autonomous driving, etc.

https://medium.com/media/cfb75faaa27327c96bdb7516644e2e58/href

#TensorFlow Tensorflow: an Introduction by ML GDE Fares Al-Qunaieer (Saudi Arabia) discussed what TensorFlow is and its uses, and the many components and tools in the framework. He also shared about the TensorFlow community and resources to learn the framework.

#TensorFlow In Visualising System Resources with TensorBoard by GDG CLOUD Islamabad, ML GDE Imran us Salam Mian (Germany) explained how to quantify the performance of an ML application using the TensorFlow Profiler.

https://medium.com/media/6761fba70b6a6e6dcff2d637c3ed8585/href

#TensorFlow Custom YOLOv7 Object Detection with TensorFlow.js by ML GDE Hugo Zanini Gomes (Brazil) is an article explaining how he trained a custom YOLOv7 model run it directly in the browser in real-time and offline with TensorFlow.js.

#TensorFlow Introduction to Tensorflow.js by TFUG Surat was to let javascript developers learn how to develop ML models using TensorFlow.js.

#TensorFlow TFUG Delhi: Exploring the Frontiers of Machine Learning by TFUG Delhi was to educate students on the new developments in AI and create more awareness about the TFUG chapters. They discussed how to develop AI projects ethically and had demos for tf.js and TF models for voice tone interpretation.

#TensorFlow Simple ML for Google Sheets (slide) by TFUG Hajipur was to educate attendees about the concept of machine learning and how Simple ML for Google Sheets can be used as a tool to leverage its power. They shared practical insights into harnessing the power of ML without the need for coding or ML expertise. In the hands-on session, participants learned how to use the tool to analyze data, make predictions, and uncover valuable insights.

#Kaggle Skeleton Based Action Recognition: A failed attempt by ML GDE Ayush Thakur (India) is a discussion post about documenting his learnings from competing in the Kaggle competition, Google — Isolated Sign Language Recognition. He shared his repository, training logs, and ideas he approached in the competition.

#Kaggle ICR | EDA & Baseline by ML GDE Ertuğrul Demir (Turkey) is a starter notebook for newcomers interested in the latest featured code competition on Kaggle. This notebook got 300+ forks.

#Cloud Document AI and AutoML — Transform and Automate your Workflows by ML GDE Sachin Kumar (Qatar) delivered how Document AI along with AutoML can be used to automatically parse existing documents and extract meaningful data and insights, as well as automate and streamline existing workflows traditionally done by humans.

#Cloud Search and AI on Google Cloud by ML GDE Gad Benram (Portugal) explored the synergy between search, AI, and Matching Engine, a vector database by GCP and its impact on industries. He showcased its capabilities in handling large-scale datasets, performing ultra-fast similarity searches, and supporting real-time data retrieval.

#Cloud In May Meetup by GDG Cloud Berlin, Googler Amrit Raj shared an exploration of enriching and transforming data using Dataform (ELT approach), training ML models using BigQuery ML, and serving/monitoring the models in Vertex AI. Another Googler Nikolai Danylchyk shared how to leverage Google’s open source tools together with Unreal Engine 5 to create global-scale multiplayer games on GCP.

[MLDP Newsletter] May2023 — Machine Learning Communities: highlights and achievements was originally published in Google Developer Experts on Medium, where people are continuing the conversation by highlighting and responding to this story.