A summary of recent updates in the field of Artificial Intelligence (AI) for week 43 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 43 of 2020

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

  • End-to-end ASR for Alexa: Amazon’s Alexa is becoming better with time due to their adoption of an end-to-end pipeline for speech recognition. This change has bypassed the typical sequence as input for acoustic or language models and moved on to the direct speech as input with transcribed output. To read the full entire visit here
  • LandingLens: The Deeplearning.ai has its newsletter for this week with key article addressed to Landing AI’s LandingLens which is an AI platform to recognize manufacturing faults in products if any. The salient features of this mainstream platform can be found here
  • Unartificial Intelligence: The November-December article release from HBR written by Scott Berinato gives a non-technical picture of the new wave of brain science that is seizing our human mind.
  • Global collaboration to implement AI principles in practice: The AI policy Forum, a broader platform for governing policies and tools for all the government and non-government bodies to implement in concrete ways. To know more, visit this link.
  • Google’s Lending DocAI: The Google Cloud has launched its first-ever tool to review documents for the mortgage companies to speed up the entire process. It basically uses the Document AI platform that offers accurate OCR. To know more details, visit here
  • Amazon’s scholars on AI ethics: The Amazon’s scholars, Kearns and Roth spoke about the journey of ML and AI so far, where it stands and the future perspective of them. In conversation with Amazon Science, these great minds had lots to say, and to find out more visit here
  • Facebooks’ Polyglot: The Polyglot AI is an open-source tool developed by Facebook for translating 100 languages amongst themselves. Here is an article mentioning the details of the same.
  • DeepMind’s ML model: The ML model is based on an old statistical model, Casual Bayesian Networks that use graph structure. To know more, visit here.
  • Why deep learning work even when it shouldn’t: The literature demands proof and not every phenomenon is explainable. A few fun facts explained here which when looked at from a logical point of view raise doubts about deep learning principles. 
  • DeepMinds open sources FermiNet: The FermiNet (Fermionic Neural Network) is a deep learning framework for computing atomic energies. The details about open-sourcing FermiNet can be found here while the Github script is available here.
  • The dark side of DeepFakes: The Sensity AI has found a publicly available bot on Telegram that converts any photo into nudes within minutes. This deep fake bot is connected to several channels on Telegram has quite a fan following amongst is members. Here is the detailed article.

Articles for week 43 of 2020

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

  • Language in-variant learning for UMT: The research post by Machine Learning Department, CMU describes the tradeoff between how well the machine translates a sentence (i.e. translation quality) to the degree of representation invariance with regard to the specific language.  Here is a blog post for a more detailed read.
  • Reverse Toonification: Nathan Shipley created photo-realistic characters for Pixar using the Pixel2Style2Pixel and to see astonishing results visit this link.
  • Linear Regression tutorial: A simple descriptive blog post on Linear Regression can be found here with python support. 
  • Increase Object Detection speed: The blog post explains the speeding of YOLO object detection FPS using transpose and reshape functions from NumPy. Here is the entire post for reference.
  • Credit card fraud detection: The article covers the aspects of preventing fraudulent attacks causing credit cardholders a lot of loss. The post explains the procedure to detect such attacks using Isolation First, Local Outlier Factor, and SVM for detection. Out of these, the Isolation First proved to be more accurate. 
  • Build a personal Voice Assistant: To build your personalized Voice assistant, the post explains the requirements, procedures, and implementation steps. Here is the entire article. 
  • Few shot-learning and Meta-learning: The new eye-catching technique that is quite useful when the training instances are limited to a few or one. The few shot-learning is based on N-way-K-shot classification and the full article can be accessed here. A couple of videos on the same are also available here.
  • Expression recognition on US presidential debate: The post covers expressions recognition on the recent attention grasped by the presidential debate. The implementation and results are also striking. 
  • CLEANN for trojans: CLEANN is a framework that protects ANN from trojans. The algorithm has achieved positive results on image classification. Here is the full article details.


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


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

  • MorphSensor: The MorphSensor is an effort by MIT researchers to design interactive objects into a physical product. Here is their demo, research article, and news coverage.
  • Spectral Centroid and Bandwidth extraction: The video shows the implementation of extracting Spectral Centroid and bandwidth using Python. Here is the demo for reference.


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

  • Altruist for IML: The Altruist is an Interpretable Machine Learning (IML) meta-learning algorithm for choosing the right features for our application using logical argumentation. To read more about Altruist, read this while the research paper and Github script are also available.
  • Faceshifter: The Faceshifter is a face-swapping algorithm that uses AIE-Net and HEAR-Net for face shifting. But this particular implementation is based on AIE-Net while the main source research article uses both techniques.
  • Batch exploration for RL: The Batch Exploration with Examples (BEE) is proposed to handle vision-based manipulation tasks on a simulator as well as a real Franka robot and proved to better than task-agnostic and weakly-supervised exploration algorithms. Read the entire article here.
  • Off-policy confidence interval estimation: The CoinDICE is a confidence interval estimation algorithm and the proposed technique is better than existing ones. Read the full article here.


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

  • ML/ Duolingo Seminar: A seminar on “Towards Embodied Intelligence” is due to be held on 27th October 2020. To register visit here
  • Multi-lingual Machine Translation: The implementation of a many-to-many language translation operation that can translate 100 languages directly. The Github code is available here.

That’s all for the week 43 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.