Data Phoenix Digest - 02.09.2021

Open Data Science Odessa Meetup #4, AI with "natural" voices from NVIDIA, deploying NVIDIA Triton at Scale with MIG and Kubernetes, self-supervised CT denoising, creating synthetic data, VIL-100, Neural-GIF, YOLOP, FedScale, AP-10K, jobs, and more...

Dmitry Spodarets

The Data Phoenix Events team is happy to remind you that in September, we will organize the following events:

Register for all events or for some specific which you like, freedom of choice is yours! We would love to see you anytime!


NEWS

What's new this week?

AI with "natural" voices from NVIDIA. AI helping biologists determine the 3-D shapes of biological molecules. Marketing AI: Subject line analysis done by a machine.

  • NVIDIA’s text-to-speech research team unveils RAD-TTS, a system that allows an individual to train a text-to-speech model with their own voice. Another feature is voice conversion.
  • Researchers from Stanford University propose an approach to determine the 3-D shapes of biological molecules by predicting accurate structures computationally, with AI/ML.
  • A new report from AI-powered copywriting platform Phrasee reveals what brand language will resonate with consumers. Phrasee analyzed millions of subject lines using AI for the report.

Funding news:

  • Peak raises a $75 million Series C funding round led by SoftBank’s Vision Fund
  • Level AI announces a $13 million Series A led by Battery Ventures
  • DiA closes a $14 million in additional investment led by a consortium of investors

ARTICLES

Deploying NVIDIA Triton at Scale with MIG and Kubernetes
In this post, you'll learn best practices for deploying multiple Triton Inference Servers with MIG and autoscaling Triton Inference Servers using Kubernetes and Prometheus monitoring stack.

DeepSpeed Powers 8x Larger MoE Model Training with High Performance
Learn about Microsoft's DeepSpeed MoE, a high-performance system that supports massive scale mixture of experts (MoE) models as part of the DeepSpeed optimization library.

SSWL-IDN: Self-Supervised CT Denoising
In this article, Ayaan Hague provides an explanation of SSWL-IDN that leverages residual learning and a hybrid loss combining perceptual loss and MSE, all incorporated in a VAE framework.

Fine-Tuning Transformer Model for Invoice Recognition
In this step-by-step guide, you'll learn how to finetune the recently released Microsoft’s Layout LM model on an annotated custom dataset, from annotation to inference.

3D Pose Detection with MediaPipe BlazePose GHUM and TensorFlow.js
In this guided walkthrough, you'll learn the specifics of TensorFlow's first 3D model in TF.js pose-detection API. The demo is by MediaPipe and TensorFlow.js.

How to Train a BERT Model From Scratch
A BERT 101. In this article, you'll find a step-by-step guide of training a functional model from scratch. All guidelines are clear and will work well for beginners.

Bootstrap a Modern Data Stack in 5 minutes with Terraform
The guide with all the details to walk you through setting up Airbyte, BigQuery, dbt, Metabase, and everything else you need to run a Modern Data Stack using Terraform.

Machine Learning Pipeline End-to-End Solution
ML implementations are supported by dozens of different services. Complexity is not necessarily a good thing. In this article, you'll learn how ML system can be split into as few services as possible.

Creating Synthetic Data for Machine Learning
This tutorial will guide you through the steps needed to create the synthetic data and show how you can then train it with YOLOv5 in order to work on real images.

PAPERS

Neural-GIF: Neural Generalized Implicit Functions for Animating People in Clothing
Neural Generalized Implicit Functions (Neural-GIF) is a method to animate people in clothing as a function of the body pose trained on various raw 3D scans.

VIL-100: A New Dataset and A Baseline Model for Video Instance Lane Detection
In the paper, the authors propose a new baseline model, named multi-level memory aggregation network (MMA-Net), for video instance lane detection.

DL-Traff: Survey and Benchmark of Deep Learning Models for Urban Traffic Prediction
In the paper, the authors review the deep traffic models and the widely used datasets, build a standard benchmark to evaluate their performances with the same settings and metrics.

YOLOP: You Only Look Once for Panoptic Driving Perception
YOLOP is a panoptic driving perception network designed to simultaneously perform traffic object detection, drivable area segmentation, and lane detection. Learn more about the model.

SwinIR: Image Restoration Using Swin Transformer
SwinIR is a strong baseline model for image restoration based on the Swin Transformer. The results demonstrate that SwinIR outperforms state-of-the-art methods on different tasks.

FedScale: Benchmarking Model and System Performance of Federated Learning
FedScale is a diverse set of challenging and realistic benchmark datasets to facilitate scalable, comprehensive, and reproducible federated learning (FL) research.

AP-10K: A Benchmark for Animal Pose Estimation in the Wild
AP-10K is the first large-scale benchmark for general animal pose estimation that is designed to facilitate the research in animal pose estimation. Learn more about it!

JOBS

Looking to feature your open positions in the digest? Kindly reach out to us at [email protected] for details. We'll be proud to help your business thrive!

Digest