Posted by Nari Yoon, Bitnoori Keum, Hee Jung, DevRel Group Supervisor / Soonson Kwon, DevRel Program Supervisor

ML Examine Jams: TFUG Bauchi, GDSC Uninter, TFUG Abidjan, MLAct, Universitas Pendidikan Indonesia, Nationwide Institute of Expertise (Kosen), Kumamoto School, GDG Assiut, GDG Bassam, GDG Cloud Abidjan, GDG Antananarivo, Madan Mohan Malaviya College of Expertise – Gorakhpur, Université d’Abomey-Calavi (UAC), ABES Engineering School – Ghaziabad, ABV-IIITM, Vishwakarma College – Pune, Pimpri Chinchwad School of Engineering and Analysis – Pune, GDG Cloud Edmonton, GDG Cocody, GDG Cloud Wilmington, College of Lay Adventist of Kigali

ML Paper Studying Golf equipment: GalsenAI, TFUG Dhaka, Pseudo Lab, TFUG Durg, TFUG Ibadan, Universidad Nacional de Ingeniería, GDG Karaganda, Western College, GDG Raipur, College School Dublin

ML Math Golf equipment: TFUG Dhaka, TFUG Hajipur, GDG Yangon, GalsenAI

Group Highlights

Keras

Picture Segmentation utilizing Composable Absolutely-Convolutional Networks by ML GDE Suvaditya Mukherjee (India) is a Kears.io instance explaining find out how to implement a fully-convolutional community with a VGG-16 backend and find out how to use it for performing picture segmentation. His presentation, KerasCV for the Younger and Stressed (slides | video) at TFUG Malaysia and TFUG Kolkata was an introduction to KerasCV. He mentioned how primary pc imaginative and prescient parts work, why Keras is a crucial device, and the way KerasCV builds on prime of the established TFX and Keras ecosystem.

[ML Story] My Keras Chronicles by ML GDE Aritra Roy Gosthipaty (India) summarized his story of stepping into deep studying with Keras. He included pointers as to how one might get into the open supply group. Plus, his Kaggle pocket book, [0.11] keras starter: unet + tf knowledge pipeline is a starter information for Vesuvius Problem. He and Subvaditya additionally shared Keras implementation of Temporal Latent Bottleneck Networks, proposed in the paper.

KerasFuse by ML GDE Ayse Ayyuce Demirbas (Portugal) is a Python library that mixes the facility of TensorFlow and Keras with varied pc imaginative and prescient strategies for medical picture evaluation duties. It gives a group of modules and features to facilitate the event of deep studying fashions in TensorFlow & Keras for duties akin to picture segmentation, classification, and extra.

TensorFlow at Google I/O 23: A Preview of the New Options and Instruments by TFUG Ibadan explored the preview of the newest options and instruments in TensorFlow. They coated a variety of matters together with Dtensor, KerasCV & KerasNLP, TF quantization API, and JAX2TF.

StableDiffusion – Textual-Inversion implementation app by ML GDE Dimitre Oliveira (Brazil) is an instance of find out how to implement code from analysis and fine-tunes it utilizing the Textual Inversion course of. It additionally gives related use circumstances for beneficial instruments and frameworks akin to HuggingFace, Gradio, TensorFlow serving, and KerasCV.

In Understanding Gradient Descent and Constructing an Picture Classifier in TF From Scratch, ML GDE Tanmay Bakshi (Canada) talked about find out how to develop a strong instinct for the basics backing ML tech, and really constructed an actual picture classification system for canine and cats, from scratch in TF.Keras.

TensorFlow and Keras Implementation of the CVPR 2023 paper by Usha Rengaraju (India) is a analysis paper implementation of BiFormer: Imaginative and prescient Transformer with Bi-Stage Routing Consideration.

Smile Detection with Python, OpenCV, and Deep Studying by Rouizi Yacine is a tutorial explaining find out how to use deep studying to construct a extra strong smile detector utilizing TensorFlow, Keras, and OpenCV.

Kaggle

ML Olympiad for College students by GDSC UNINTER was for college kids and aspiring ML practitioners who need to enhance their ML abilities. It consisted of a problem of predicting US working visa functions. 320+ attendees registered for the opening occasion, 700+ views on YouTube, 66 groups competed, and the winner received a 71% F1-score.

ICR | EDA & Baseline by ML GDE Ertuğrul Demir (Turkey) is a starter pocket book for newcomers within the newest featured code competitors on Kaggle. It received 200+ Upvotes and 490+ forks.

Compete Extra Successfully on Kaggle utilizing Weights and Biases by TFUG Hajipur was a meetup to discover strategies utilizing Weights and Biases to enhance mannequin efficiency in Kaggle competitions. Usha Rengaraju (India) joined as a speaker and delivered her insights on Kaggle and techniques to win competitions. She shared ideas and methods and demonstrated find out how to arrange a W&B account and find out how to combine with Google Colab and Kaggle.

Skeleton Based mostly Motion Recognition: A failed try by ML GDE Ayush Thakur (India) is a dialogue publish about documenting his learnings from competing within the Kaggle competitors, Google – Remoted Signal Language Recognition. He shared his repository, coaching logs, and concepts he approached within the competitors. Plus, his article Keras Dense Layer: Methods to Use It Accurately) explored what the dense layer in Keras is and the way it works in apply.

On-device ML

Add Machine Studying to your Android App by ML GDE Pankaj Rai (India) at Tech Talks for Educators was a session on on-device ML and find out how to add ML capabilities to Android apps akin to object detection and gesture detection. He defined capabilities of ML Equipment, MediaPipe, TF Lite and find out how to use these instruments. 700+ individuals registered for his speak.

In MediaPipe with a little bit of Bard at I/O Prolonged Singapore 2023, ML GDE Martin Andrews (Singapore) shared how MediaPipe matches into the ecosystem, and confirmed 4 completely different demonstrations of MediaPipe performance: audio classification, facial landmarks, interactive segmentation, and textual content classification.

Including ML to our apps with Google ML Equipment and MediaPipe by ML GDE Juan Guillermo Gomez Torres (Bolivia) launched ML Equipment & MediaPipe, and the advantages of on-device ML. In Startup Academy México (Google for Startups), he shared find out how to improve the worth for purchasers with ML and MediaPipe.

LLM

Introduction to Google’s PaLM 2 API by ML GDE Hannes Hapke (United States) launched find out how to use PaLM2 and summarized main benefits of it. His one other article The function of ML Engineering within the time of GPT-4 & PaLM 2 explains the function of ML consultants to find the proper steadiness and alignment amongst stakeholders to optimally navigate the alternatives and challenges posed by this rising expertise. He did shows below the identical title at North America Join 2023 and the GDG Portland occasion.

ChatBard : An Clever Buyer Service Middle App by ML GDE Ruqiya Bin Safi (Saudi Arabia) is an clever customer support middle app powered by generative AI and LLMs utilizing PaLM2 APIs.

Bard can now code and put that code in Colab for you by ML GDE Sam Witteveen (Singapore) confirmed how Bard makes code. He runs a Youtube channel exploring ML and AI, with playlists akin to Generative AI, Paper Opinions, LLMs, and LangChain.

Google’s Bard Can Write Code by ML GDE Bhavesh Bhatt (India) reveals the coding capabilities of Bard, find out how to create a 2048 sport with it, and find out how to add some primary options to the sport. He additionally uploaded movies about LangChain in a playlist and launched Google Cloud’s new course on Generative AI in this video.

Consideration Mechanisms and Transformers by GDG Cloud Saudi talked about Consideration and Transformer in NLP and ML GDE Ruqiya Bin Safi (Saudi Arabia) participated as a speaker. One other occasion, Arms-on with the PaLM2 API to create sensible apps(Jeddah) explored what LLMs, PaLM2, and Bard are, find out how to use PaLM2 API, and find out how to create sensible apps utilizing PaLM2 API.

Arms-on with Generative AI: Google I/O Prolonged [Virtual] by ML GDE Henry Ruiz (United States) and Net GDE Rabimba Karanjai (United States) was a workshop on generative AI exhibiting hands-on demons of find out how to get began utilizing instruments akin to PaLM API, Hugging Face Transformers, and LangChain framework.

Generative AI with Google PaLM and MakerSuite by ML GDE Kuan Hoong (Malaysia) at Google I/O Prolonged George City 2023 was a speak about LLMs with Google PaLM and MakerSuite. The occasion hosted by GDG George City and likewise included ML matters akin to LLMs, accountable AI, and MLOps.

Intro to Gen AI with PaLM API and MakerSuite by TFUG São Paulo was for individuals who need to study generative AI and the way Google instruments may help with adoption and worth creation. They coated find out how to begin prototyping Gen AI concepts with MakerSuite and find out how to entry superior options of PaLM2 and PaLM API. The group additionally hosted Opening Pandora’s field: Understanding the paper that revolutionized the sector of NLP (video) and ML GDE Pedro Gengo (Brazil) and ML GDE Vinicius Caridá (Brazil) shared the key behind the well-known LLM and different Gen AI fashions.The group members studied Consideration Is All You Want paper collectively and discovered the complete potential that the expertise can provide.

Language fashions which PaLM can converse, see, transfer, and perceive by GDG Cloud Taipei was for individuals who need to perceive the idea and software of PaLM. ML GED Jerry Wu (Taiwan) shared the PaLM’s predominant traits, features, and and so forth.

Serving With TF and GKE: Steady Diffusion by ML GDE Chansung Park (Korea) and ML GDE Sayak Paul (India) discusses how TF Serving and Kubernetes Engine can serve a system with on-line deployment. They broke down Steady Diffusion into predominant parts and the way they affect the next consideration for deployment. Then additionally they coated the deployment-specific bits akin to TF Serving deployment and k8s cluster configuration.

TFX + W&B Integration by ML GDE Chansung Park (Korea) reveals how KerasTuner can be utilized with W&B’s experiment monitoring function inside the TFX Tuner part. He developed a customized TFX part to push a full-trained mannequin to the W&B Artifact retailer and publish a working software on Hugging Face House with the present model of the mannequin. Additionally, his speak titled, ML Infra and Excessive Stage Framework in Google Cloud Platform, delivered what MLOps is, why it’s exhausting, why cloud + TFX is an efficient starter, and the way TFX is seamlessly built-in with Vertex AI and Dataflow. He shared use circumstances from the previous initiatives that he and ML GDE Sayak Paul (India) have accomplished within the final 2 years.

Open and Collaborative MLOps by ML GDE Sayak Paul (India) was a speak about why openness and collaboration are two vital facets of MLOps. He gave an summary of Hugging Face Hub and the way it integrates properly with TFX to advertise openness and collaboration in MLOps workflows.

ML Analysis

Paper evaluate: PaLM 2 Technical Report by ML GDE Grigory Sapunov (UK) appeared into the small print of PaLM2 and the paper. He shares opinions of papers associated to Google and DeepMind by way of his social channels and listed below are a few of them: Mannequin analysis for excessive dangers (paper), Quicker sorting algorithms found utilizing deep reinforcement studying (paper), Energy-seeking will be possible and predictive for skilled brokers (paper).

Studying JAX in 2023: Half 3 — A Step-by-Step Information to Coaching Your First Machine Studying Mannequin with JAX by ML GDE Aritra Roy Gosthipaty (India) and ML GDE Ritwik Raha (India) reveals how JAX can practice linear and nonlinear regression fashions and the utilization of PyTrees library to coach a multilayer perceptron mannequin. As well as, at Could 2023 Meetup hosted by TFUG Mumbai, they gave a chat titled Decoding Finish to Finish Object Detection with Transformers and coated the structure of the mode and the assorted parts that led to DETR’s inception.

20 steps to coach a deployed model of the GPT mannequin on TPU by ML GDE Jerry Wu (Taiwan) shared find out how to use JAX and TPU to coach and infer Chinese language question-answering knowledge.

Multimodal Transformers – Customized LLMs, ViTs & BLIPs by TFUG Singapore checked out what fashions, techniques, and strategies have come out not too long ago associated to multimodal duties. ML GDE Sam Witteveen (Singapore) appeared into varied multimodal fashions and techniques and how one can construct your personal with the PaLM2 Mannequin. In June, this group invited Blaise Agüera y Arcas (VP and Fellow at Google Analysis) and shared the Cerebra challenge and the analysis happening at Google DeepMind together with the present and future developments in generative AI and rising traits.

TensorFlow

Coaching a advice mannequin with dynamic embeddings by ML GDE Thushan Ganegedara (Australia) explains find out how to construct a film recommender mannequin by leveraging TensorFlow Recommenders (TFRS) and TensorFlow Recommenders Addons (TFRA). The first focus was to point out how the dynamic embeddings supplied within the TFRA library can be utilized to dynamically develop and shrink the scale of the embedding tables within the advice setting.

How I constructed probably the most environment friendly deepfake detector on this planet for $100 by ML GDE Mathis Hammel (France) was a chat exploring a technique to detect photos generated through ThisPersonDoesNotExist.com and even a approach to know the precise time the photograph was produced. Plus, his Twitter thread, OSINT Investigation on LinkedIn, investigated a community of pretend corporations on LinkedIn. He used a selfmade device primarily based on a TensorFlow mannequin and hosted it on Google Cloud. Technical explanations of generative neural networks had been additionally included. Greater than 701K individuals seen this thread and it received 1200+ RTs and 3100+ Likes.

Few-shot studying: Making a real-time object detection utilizing TensorFlow and Python by ML GDE Hugo Zanini (Brazil) reveals find out how to take footage of an object utilizing a webcam, label the pictures, and practice a few-shot studying mannequin to run in real-time. Additionally, his article, Customized YOLOv7 Object Detection with TensorFlow.js explains how he skilled a customized YOLOv7 mannequin to run it straight within the browser in actual time and offline with TensorFlow.js.

The Lord of the Phrases : The Return of the experiments with DVC (slides) by ML GDE Gema Parreno Piqueras (Spain) was a chat explaining Transformers within the neural machine studying situation, and find out how to use Tensorflow and DVC. Within the challenge, she used Tensorflow Datasets translation catalog to load knowledge from varied languages, and TensorFlow Transformers library to coach a number of fashions.

Speed up your TensorFlow fashions with XLA (slides) and Ship quicker TensorFlow fashions with XLA by ML GDE Sayak Paul (India) shared find out how to speed up TensorFlow fashions with XLA in Cloud Group Days Kolkata 2023 and Cloud Group Days Pune 2023.

Setup of NVIDIA Merlin and Tensorflow for Advice Fashions by ML GDE Rubens Zimbres (Brazil) introduced a evaluate of advice algorithms in addition to the Two Towers algorithm, and setup of NVIDIA Merlin on premises and on Vertex AI.

Cloud

AutoML pipeline for tabular knowledge on VertexAI in Go by ML GDE Paolo Galeone (Italy) delved into the event and deployment of tabular fashions utilizing VertexAI and AutoML with Go, showcasing the precise Go code and sharing insights gained by way of trial & error and in depth Google analysis to beat documentation limitations.

Past photos: looking out data in movies utilizing AI (slides) by ML GDE Pedro Gengo (Brazil) and ML GDE Vinicius Caridá (Brazil) confirmed find out how to create a search engine the place you’ll be able to seek for data in movies. They introduced an structure the place they transcribe the audio and caption the frames, convert this textual content into embeddings, and save them in a vector DB to have the ability to search given a consumer question.

The key sauce to creating superb ML experiences for builders by ML GDE Gant Laborde (United States) was a podcast sharing his “aha” second, 20 years of expertise in ML, and the key to creating pleasant and significant experiences for builders.

What’s inside Google’s Generative AI Studio? by ML GDE Gad Benram (Portugal) shared the preview of the brand new options and what you’ll be able to anticipate from it. Moreover, in Methods to pitch Vertex AI in 2023, he shared the six easy and trustworthy gross sales pitch factors for Google Cloud representatives on find out how to persuade prospects that Vertex AI is the proper platform.

In Methods to construct a conversational AI Augmented Actuality Expertise with Sachin Kumar, ML GDE Sachin Kumar (Qatar) talked about find out how to construct an AR app combining a number of applied sciences like Google Cloud AI, Unity, and and so forth. The session walked by way of the step-by-step means of constructing the app from scratch.

Machine Studying on Google Cloud Platform by ML GDE Nitin Tiwari (India) was a mentoring aiming to offer college students with an in-depth understanding of the processes concerned in coaching an ML mannequin and deploying it utilizing GCP. In Constructing strong ML options with TensorFlow and GCP, he shared find out how to leverage the capabilities of GCP and TensorFlow for ML options and deploy customized ML fashions.

Knowledge to AI on Google cloud: Auto ML, Gen AI, and extra by TFUG Prayagraj educated college students on find out how to leverage Google Cloud’s superior AI applied sciences, together with AutoML and generative AI.