6342
6342
6342
6342
6342 Posted by Nari Yoon, Hee 6342 Jung, DevRel Group Supervisor / 6342 Soonson Kwon, DevRel Program Supervisor
6342
6342
Let’s discover highlights and accomplishments 6342 of huge Google Machine Studying 6342 communities over the second quarter 6342 of the 12 months! We’re 6342 enthusiastic and grateful about all 6342 of the actions by the 6342 worldwide community of ML communities. 6342 Listed here are the highlights!
6342
6342 TensorFlow/Keras
6342
6342
6342 TFUG Agadir 6342 hosted 6342 #MLReady 6342 section as part of 6342 #30DaysOfML. #MLReady aimed to arrange 6342 the attendees with the information 6342 required to know the various 6342 kinds of issues which deep 6342 studying can resolve, and helped 6342 attendees be ready for the 6342 TensorFlow Certificates.
6342
6342
6342
6342 TFUG Taipei 6342 hosted the fundamental Python 6342 and TensorFlow programs named 6342 From Python to TensorFlow 6342 . The goal of those 6342 occasions is to assist everybody 6342 be taught in regards to 6342 the fundamentals of Python and 6342 TensorFlow, together with TensorFlow Hub, 6342 TensorFlow API. The occasion movies 6342 are shared each week by 6342 way of 6342 Youtube playlist 6342 .
6342
6342
6342
6342 TFUG New York 6342 hosted 6342 Introduction to Neural Radiance Fields 6342 for TensorFlow customers. The 6342 speak included Quantity Rendering, 3D 6342 view synthesis, and hyperlinks to 6342 a minimal implementation of NeRF 6342 utilizing Keras and TensorFlow. Within 6342 the occasion, ML GDE Aritra 6342 Roy Gosthipaty (India) had a 6342 chat specializing in breaking the 6342 ideas of the educational paper, 6342 6342 NeRF: Representing Scenes as Neural 6342 Radiance Fields for View Synthesis 6342 into easier and extra 6342 ingestible snippets.
6342
6342
6342 TFUG Turkey 6342 , GDG Edirne and GDG 6342 Mersin organized a 6342 TensorFlow Bootcamp 22 6342 and ML GDE M. 6342 Yusuf Sarıgöz (Turkey) participated as 6342 a speaker, 6342 TensorFlow Ecosystem: Get most out 6342 of auxiliary packages 6342 . Yusuf demonstrated the inside 6342 workings of TensorFlow, how variables, 6342 tensors and operations work together 6342 with one another, and the 6342 way auxiliary packages are constructed 6342 upon this skeleton.
6342
6342
6342
6342 TFUG Mumbai 6342 hosted the 6342 June Meetup 6342 and 110 of us 6342 gathered. ML GDE Sayak Paul 6342 (India) and TFUG mentor Darshan 6342 Despande shared information by way 6342 of classes. And ML workshops 6342 for novices went on and 6342 members constructed up machine studying 6342 fashions with out writing a 6342 single line of code.
6342
6342
6342
ML GDE Hugo Zanini (Brazil) 6342 wrote 6342 Realtime SKU detection within the 6342 browser utilizing TensorFlow.js 6342 . He shared an answer 6342 for a well known downside 6342 within the client packaged items 6342 (CPG) business: real-time and offline 6342 6342 SKU 6342 detection utilizing TensorFlow.js.
6342
6342
ML GDE Gad Benram (Portugal) 6342 wrote 6342 Can a pair TensorFlow strains 6342 scale back overfitting? 6342 He defined how just 6342 some strains of code can 6342 generate knowledge augmentations and enhance 6342 a mannequin’s efficiency on the 6342 validation set.
6342
6342
ML GDE Victor Dibia (USA) 6342 wrote 6342 The way to Construct An 6342 Android App and Combine Tensorflow 6342 ML Fashions 6342 sharing how one can 6342 run machine studying fashions regionally 6342 on Android cellular gadgets, 6342 The way to Implement Gradient 6342 Explanations for a HuggingFace Textual 6342 content Classification Mannequin (Tensorflow 2.0) 6342 explaining in 5 steps 6342 about how one can confirm 6342 the mannequin is specializing in 6342 the suitable tokens to categorise 6342 textual content. He additionally wrote 6342 6342 how one can finetune a 6342 HuggingFace mannequin for textual content 6342 classification, utilizing Tensorflow 2.0 6342 .
6342
6342
ML GDE Karthic Rao (India) 6342 launched a brand new sequence 6342 6342 ML for JS builders with 6342 TFJS 6342 . This sequence is a 6342 mixture of brief portrait and 6342 lengthy panorama movies. You’ll be 6342 able to discover ways to 6342 construct a poisonous phrase detector 6342 utilizing TensorFlow.js.
6342
6342
ML GDE Sayak Paul (India) 6342 applied the DeiT household of 6342 ViT fashions, ported the pre-trained 6342 params into the implementation, and 6342 offered code for off-the-shelf inference, 6342 fine-tuning, visualizing consideration rollout plots, 6342 distilling ViT fashions by way 6342 of consideration. ( 6342 code 6342 | 6342 pretrained mannequin 6342 | 6342 tutorial 6342 )
6342
6342
6342
ML GDE Sayak Paul (India) 6342 and ML GDE Aritra Roy 6342 Gosthipaty (India) inspected varied phenomena 6342 of a Imaginative and prescient 6342 Transformer, shared insights from varied 6342 related works executed within the 6342 space, and offered concise implementations 6342 which might be appropriate with 6342 Keras fashions. They supply instruments 6342 to probe into the representations 6342 discovered by completely different households 6342 of Imaginative and prescient Transformers. 6342 ( 6342 tutorial 6342 | 6342 code 6342 )
6342
6342 JAX/Flax
6342
6342
6342
ML GDE Aakash Nain (India) 6342 had a particular speak, 6342 Introduction to JAX 6342 for ML GDEs, TFUG 6342 organizers and ML neighborhood community 6342 organizers. He coated the basics 6342 of JAX/Flax in order that 6342 increasingly more folks check out 6342 JAX within the close to 6342 future.
6342
6342
ML GDE Seunghyun Lee (Korea) 6342 began a undertaking, 6342 Coaching and Lightweighting Cookbook in 6342 JAX/FLAX 6342 . This undertaking makes an 6342 attempt to construct a neural 6342 community coaching and lightweighting cookbook 6342 together with three sorts of 6342 lightweighting options, i.e., information distillation, 6342 filter pruning, and quantization.
6342
6342
ML GDE Yucheng Wang (China) 6342 wrote 6342 Historical past and options of 6342 JAX 6342 and defined the distinction 6342 between JAX and Tensorflow.
6342
6342
6342
ML GDE Martin Andrews (Singapore) 6342 shared a video, 6342 Sensible JAX : Utilizing Hugging 6342 Face BERT on TPUs 6342 . He reviewed the Hugging 6342 Face BERT code, written in 6342 JAX/Flax, being fine-tuned on Google’s 6342 Colab utilizing Google TPUs. ( 6342 Pocket book 6342 for the video)
6342
6342
ML GDE Soumik Rakshit (India) 6342 wrote 6342 Implementing NeRF in JAX 6342 . He makes an attempt 6342 to create a minimal implementation 6342 of 3D volumetric rendering of 6342 scenes represented by Neural Radiance 6342 Fields.
6342
6342 Kaggle
6342
6342
ML GDEs’ Kaggle notebooks have 6342 been introduced because the winner 6342 of 6342 Google OSS Professional Prize 6342 on Kaggle 6342 : Sayak Paul and Aritra 6342 Roy Gosthipaty’s 6342 Masked Picture Modeling with Autoencoders 6342 in March; Sayak Paul’s 6342 6342 Distilling Imaginative and prescient Transformers 6342 in April; Sayak Paul 6342 & Aritra Roy Gosthipaty’s 6342 Investigating Imaginative and prescient Transformer 6342 Representations 6342 ; Soumik Rakshit’s 6342 Tensorflow Implementation of Zero-Reference Deep 6342 Curve Estimation 6342 in Might and Aakash 6342 Nain’s 6342 The Definitive Information to Augmentation 6342 in TensorFlow and JAX 6342 in June.
6342
6342
6342
ML GDE Luca Massaron (Italy) 6342 printed 6342 The Kaggle E-book 6342 with Konrad Banachewicz. This 6342 guide particulars competitors evaluation, pattern 6342 code, end-to-end pipelines, finest practices, 6342 and suggestions & tips. And 6342 in 6342 the net occasion 6342 , Luca and the co-author 6342 talked about how one can 6342 compete on Kaggle.
6342
6342
6342 ML GDE Ertuğrul Demir (Turkey) 6342 wrote 6342 Kaggle Handbook: Fundamentals to Survive 6342 a Kaggle Shake-up 6342 masking bias-variance tradeoff, validation 6342 set, and cross validation method. 6342 Within the 6342 second submit 6342 of the sequence, he 6342 confirmed extra strategies utilizing analogies 6342 and case research.
6342
6342
6342
6342 TFUG Chennai 6342 hosted 6342 ML Examine Jam with Kaggle 6342 and created research teams 6342 for the members. Greater 6342 than 60% of members have 6342 been energetic throughout the entire 6342 program and lots of of 6342 them shared their completion certificates.
6342
6342
6342
TFUG Mysuru organizer 6342 Usha Rengaraju 6342 shared a 6342 Kaggle pocket book 6342 which incorporates the implementation 6342 of the analysis paper: UNETR 6342 – Transformers for 3D Biomedical 6342 Picture Segmentation. The mannequin routinely 6342 segments the abdomen and intestines 6342 on MRI scans.
6342
6342 TFX
6342
6342
6342
ML GDE Sayak Paul (India) 6342 and ML GDE Chansung Park 6342 (Korea) shared how one can 6342 deploy a deep studying mannequin 6342 with Docker, Kubernetes, and Github 6342 actions, with two promising methods 6342 – 6342 FastAPI 6342 (for REST) and 6342 TF Serving 6342 (for gRPC).
6342
6342
ML GDE Ukjae Jeong (Korea) 6342 and ML Engineers at Karrot 6342 Market, a cellular commerce unicorn 6342 with 23M customers, wrote 6342 Why Karrot Makes use of 6342 TFX, and The way to 6342 Enhance Productiveness on ML Pipeline 6342 Improvement 6342 .
6342
6342
ML GDE Jun Jiang (China) 6342 had a 6342 speak 6342 introducing the idea of 6342 MLOps, the production-level end-to-end options 6342 of Google & TensorFlow, and 6342 how one can use TFX 6342 to construct the search and 6342 advice system & scientific analysis 6342 platform for large-scale machine studying 6342 coaching.
6342
6342
6342
ML GDE Piero Esposito (Brazil) 6342 wrote 6342 Constructing Deep Studying Pipelines with 6342 Tensorflow Prolonged 6342 . He confirmed how one 6342 can get began with TFX 6342 regionally and how one can 6342 transfer a TFX pipeline from 6342 native setting to Vertex AI; 6342 and offered code samples to 6342 adapt and get began with 6342 TFX.
6342
6342
6342
6342 TFUG São Paulo 6342 (Brazil) had a sequence 6342 of on-line webinars on TensorFlow 6342 and TFX. Within the TFX 6342 session, they centered on how 6342 one can put the fashions 6342 into manufacturing. They talked in 6342 regards to the knowledge buildings 6342 in TFX and implementation of 6342 the primary pipeline in TFX: 6342 ingesting and validating knowledge.
6342
6342
TFUG Stockholm hosted 6342 MLOps, TensorFlow in Manufacturing, and 6342 TFX 6342 masking why, what and 6342 how one can successfully leverage 6342 MLOps finest practices to scale 6342 ML efforts and had a 6342 have a look at how 6342 TFX can be utilized for 6342 designing and deploying ML pipelines.
6342
6342 Cloud AI
6342
6342
6342
ML GDE Chansung Park (Korea) 6342 wrote 6342 MLOps System with AutoML and 6342 Pipeline in Vertex AI 6342 on GCP official weblog. 6342 He confirmed how Google Cloud 6342 Storage and Google Cloud Capabilities 6342 will help handle knowledge and 6342 deal with occasions within the 6342 MLOps system.
6342
6342
He additionally shared the Github 6342 repository, 6342 Steady Adaptation with VertexAI’s AutoML 6342 and Pipeline 6342 . This incorporates two notebooks 6342 to exhibit how one can 6342 automate to provide a brand 6342 new AutoML mannequin when the 6342 brand new dataset is available 6342 in.
6342
6342
TFUG Northwest (Portland) hosted 6342 The State and Way forward 6342 for AI + ML/MLOps/VertexAI lab 6342 walkthrough 6342 . On this occasion, ML 6342 GDE Al Kari (USA) outlined 6342 the know-how panorama of AI, 6342 ML, MLOps and frameworks. Googler 6342 Andrew Ferlitsch had a speak 6342 about Google Cloud AI’s definition 6342 of the 8 levels of 6342 MLOps for enterprise scale manufacturing 6342 and the way Vertex AI 6342 suits into every stage. And 6342 MLOps engineer Chris Thompson coated 6342 how straightforward it’s to deploy 6342 a mannequin utilizing the Vertex 6342 AI instruments.
6342
6342 Analysis
6342
6342
6342
ML GDE Qinghua Duan (China) 6342 launched a 6342 video 6342 which introduces Google’s newest 6342 540 billion parameter mannequin. He 6342 launched the paper PaLM, and 6342 described the fundamental coaching course 6342 of and improvements.
6342
6342
ML GDE Rumei LI (China) 6342 wrote weblog postings reviewing papers, 6342 6342 DeepMind’s Flamingo 6342 and 6342 Google’s PaLM 6342 .
6342
6342