A summary of recent updates in the field of Artificial Intelligence (AI) for week 42 of 2020. For more regular updates and related discussions, join us in the Artificial Intelligence Community. As always, thanks for your support for sharing it with your friends and colleagues on social media.

Industry updates for week 42 of 2020

This section is for recent industry updates related to Artificial Intelligence.

  • Microsoft’s effort to overcome data shortage for disability: The gadgets and technology built for people with disability have limited data for training and hence do not work well an are not very user friendly. Microsoft along with its non-profit organization is trying to make these tools better for people with blindness and partial disability. Here is the source article.
  • Zentane and ONNX’s joint venture: Zentane Systems have joined the open-source community, ONNX, with the growing need to develop open innovating products. The ONNX with Zentane aims to lend companies with services related to AI. To know more, read here.
  • Signature of Lion’s Roar: The researchers at Oxford University have installed biologgers that capture the movement and the roar of the individual lion and identified a way to distinguish them from one another using a pattern recognition algorithm. To read the entire article, visit this link.
  • Deep Fakes, the next social media hype: This year Deep fakes have been on the internet news for every reason and almost every week it has managed to raise concerns with the growing instances of crimes. So far 14000 plus deep videos are available online. To read the full article, visit here.
  • Google’s new search feature with eyes and ears: Google has released a statement saying its users would be able to find out crowded places using its Maps and also if people are following safety precautions for Covid19. Along with this, its spell check feature in text and video highlighting features are also on their table for making users experience better through enhanced NLP and AR respectively. To read the entire article, visit this link.
  • NVIDIA to support Leonardo: The Italian Supercomputer Leonardo will have NVIDIA exaflop with FP16. Along with this, it will also provide Graphics card Mellanox HDR 200GB/s Infiniband networking. To get more details, visit here.
  • Future AI generations: The growth of AI is drastic with just a few years of the generation gap between the newest algorithms. The article covers three areas where AI is bound to emerge in the community.
  • NVIDIA’s opensource conversational AI model: This comes after NVIDIA announced the NeMo model just last week which is. PyTorch based open-source platform for speech and language models. The benefits of NeMo are mentioned here in this article.
  • NeurIPS faces bad publicity due to “Terrible” reviews: This year’s NeurIPS 2020 is due to be held in December second week but is all over the tech news already. The reason mentioned by an AI researcher from NVIDIA is incomplete or less detailed review comments. Here is the entire news report. 
  • Facebook’s Graph-Based ML: Facebook’s open-source Graph Transformer Networks (GTN) are effective and are explained wrt the automatic differentiation of WFSTs in this article.
  • Open Catalyst Project: Facebook and CMU has launched Open Catalyst Project to improvise and provide a pace to quantum mechanical simulations by 1k times. This venture intends to promote discovering new catalysts meant for storing and usage of renewable energy. Here is the entire article for a quick read.

Articles for week 42 of 2020

This section is for recent Blog Posts/Tutorials related to Artificial Intelligence.

  • AI internet: AI is everywhere and so is the internet. But a smart internet is an adding recommendation system, personalized suggestions, image, speech, text-based searches, you name it and it’s there on the internet. This article explains the technological benefits of the smart world around us through AI.
  • Object Detection Models: The popular object detection models such as the R-CNN and YOLO are evaluated using mAP based evaluation measures. The Mean average precision is the score of ground truth bounding box wrt detected box. The higher the score the better is the accuracy of the model. To know more, visit here.
  • Singularity may not require AGI: Singularity is a topic of discussion amongst researchers for quite some time now. The article here opens a new question about the need for Artificial General Intelligence (AGI) to attain Singularity. AGI is nowhere close to development and the AI engineers still don’t know where to start from.
  • Reconstructing streetscapes using ML: To get a time travel experience to see how our favorite places look like in the past, Google has built a toolkit, rǝ (re’ turn) using Kubernetes. The tool is based on three components, namely crowd-sourced platform, temporal map server, and 3D experience platform. Here is the entire article.
  • Less Than One-Shot Learning: The state-of-the-art ML algorithms need tons of data for training to learn the pattern in the data. The Less-than-One Shot learning can identify more objects than it is trained on. To get more information, visit here.
  • Reinforcement Learning is supervised learning: Reinforcement learning can be seen as using supervised learning with optimized data. This blog post shows an alternate perspective of applying RL.
  • The Canonical Stack for ML: A well-written post on AI emerging on a progressive basis with every problem being solved and opening an innovation agenda. The state of events in AI writes from Facebook to Fiddler, each tech giant is building his own AI and the attackers are building stronger AI to get over them. And the competition goes on. Read the entire article here.
  • End-to-End Text Recognition: This article covers the installation of OpenVINO for Text Recognition. Besides, it also portrays a demo on CPU, iGPU, and VPU.
  • Setup Virtual Server on Cloud: This article gives a kick start to get a virtual server running on the IBM cloud. A step by step guide to build your GPU compute instance in the cloud. 
  • Neural Structured Learning: The NSL is a Tensorflow based platform to train structured data signals using Neural networks. The post mentions quite some details about how NSL handles data structures along with an implementation description.
  • Building a basic QR scannerThis is a basic project using Python’s Pyzbar library for building a barcode and QR code scanner.

Softwares

This section is for recent Tools/Applications related to Artificial Intelligence.

  • Haven AI: It’s a library built for visualizing larger-scale experiments. Here is the GitHub code.

Videos

This section is for recent Lectures/Conferences/Webinars related to Artificial Intelligence.

  • AI for Beginners – Number Recognition: For anyone who wishes to start with AI, the video is the most basic practical application of AI.
  • Future of Payment through Cards: A small glimpse of how the future of card payments would look like through the use of AR. The demo can be seen here.

Research

This section is for recent research published related to Artificial Intelligence.

  • Guide to build Graph Neural Networks: The Graph Neural Networks are gaining importance due to their ability to process unstructured data. This article helps to build a Graph Neural Network with a detailed mathematical description along with the research paper.
  • FACT diagnosisFACT is a diagnostic tool that handles the trade-offs between-group fairness. The article is mainly focused on LAFOP and the research paper as well as the code is available.
  • Speech Separation through localization: In a multi-microphone scenario, separating speakers who are speaking parallells is the task of speaker separation or diarization. The proposed technique uses a deep network in the time domain by localizing speakers. Here is a research article for an overview.

Miscellaneous

This section is for updates for other various related topics to Artificial Intelligence.

  • Knowledge-based vs AI and ML: This question might have popped up in your mind as to how is the knowledge-based approach different from general AI approaches and also its relation to ML. Here is a quick read to answer the preceding question. 
  • Multi-modal Emotion Recognition: This multi-modal emotion recognition technique uses transformers based self-supervised feature fusion. Here is the GitHub code for the same.

That’s all for the week 42 of 2020. In case you think we have missed out something and that is significant to add, please free to reach out and submit to us via our Contact form or via Social media channels: Twitter and Facebook.