AI Scholar Weekly - Issue #41



Subscribe to our newsletter

By subscribing, you agree with Revue’s Terms of Service and Privacy Policy and understand that AI Scholar Weekly will receive your email address.

AI Scholar Weekly - Issue #41
By Educate AI  • Issue #41 • View online
Google Reveals Major Hidden Weakness in ML; Interactive Exploration and Analysis of NLP Models; Gartner Top 10 Trends in Data and Analytics for 2020; Bag of Tricks for Adversarial Training; How to Detect Hateful Memes (and more)

AI Success Cases
A global professional services firm used AI to quickly and accurately spot potentially non-compliant activities
Goal: Analyze compliance and financial risks in real time
The Challenge: International due diligence involves exhaustive research. More than 40,000 global sources track not only media but also corporate records, financial transactions and legal cases. Results based on analysts’ text strings must be painstakingly reviewed for each entity before a report can be finalized. The process is laborious; it can take weeks.
Results: The firm implemented a machine learning solution  based on IBM’s Watson Explorer. The solution automates the search process, integrates the research workflow, and cuts the time spent by analysts in manually reviewing irrelevant material by more than half. It indexes information sources to report legal and/or fraud risk associated with entities of interest, analyzes risk in real time and highlights compliance risks. Its more accurate content analytics provide deeper insights into entities researched. The results in numbers; 
  • Sharply decreases researchers’ time, with 14% of reports completed in one hour. 
  • Allows the client to generate up to 30% more due diligence reports a year.
  • Performs exhaustive research on over 40,000 global sources. 
  • Allows the client to analyze compliance and financial risks in real time.
If you have a similar business need or want similar success, sign up for a 30-minute consultation now and learn how AI can upgrade your business today 
Top AI Research This Week
#1  The Language Interpretability Tool (LIT): Interactive Exploration and Analysis of NLP Models
As natural language processing (NLP) models become more powerful and are deployed in more real-world contexts, understanding their behavior is becoming increasingly critical. 
In 2018, Google AI released What-If Tool to address the challenge of enabling black-box probing of classification and regression models, which enabled researchers to more easily debug performance and analyze the fairness of machine learning models through interaction and visualization. But there was still a need for a toolkit that would address challenges specific to NLP models.
Google AI has now released, and open-sourced Language Interpretability Tool (LIT), an interactive platform for NLP model understanding. LIT is built upon the lessons learned from the What-If Tool with greatly expanded capabilities and is for researchers and practitioners looking to understand NLP model behavior through a visual, interactive, and extensible tool. LIT demos using pre-trained models available here. Read more: Interactive Exploration and Analysis of NLP Models with LIT
#2 DeepMind Open-sourced Lab2D 
DeepMind open-sourced Lab2D a platform designed to support the development of 2D environments for AI and machine learning research.
The scalable environment simulator for artificial intelligence facilitates researcher-led experimentation with environment design. DeepMind Lab2D was built with the specific needs of multi-agent deep reinforcement learning researchers in mind, but it may also be useful beyond that particular subfield. Lab2D supports multiple players interacting in the corresponding environment simultaneously. These players can be both human and computer-controlled. 
The computationally intensive engine is written in C++ for efficiency, while most of the level-specific logic is scripted in Lua. According to DeepMind, Lab2 facilitates researcher creativity in the design of learning environments and intelligence tests. They  are excited to see what the research community uses it to build in the future.  Access on Github Read the full paper: A Customisable 2D Platform for Agent-based AI Research
#3 Multimodal Learning for Hateful Memes Detection
Hateful memes spreading hatred through social networks are on the increase. Automatically detecting the hateful memes would help reduce their harmful societal influence. 
To this regard, this paper proposes a new triplet-relation module for hateful memes prediction. The model exploits the combination of image captions and memes to enhance multimodal modeling. Researchers envision such a tripletrelation network to be extended to other frameworks that require strong attention from multimodal signals. The framework achieves competitive results for the challenge of hateful memes detection. It can  be used for filtering some of the memes distributed through a social network and reduce the need for human moderators. Read more: Multimodal Learning for Hateful Memes Detection
#4 Bag of Tricks for Adversarial Training
Adversarial training (AT) is one of the most effective strategies for promoting model robustness. However, recent benchmarks show that most of the proposed improvements on AT are less effective than simply early stopping the training procedure. This counter-intuitive fact motivated researchers to investigate the implementation details of tens of AT methods. Surprisingly, they found  that the basic settings such as  weight decay, training schedule, etc. used in these methods are highly inconsistent. 
In this paper, they provide comprehensive evaluations on CIFAR-10, focusing on the effects of mostly overlooked training tricks and hyperparameters for adversarially trained models. Their empirical observations suggest that adversarial robustness is much more sensitive to some basic training settings than previously thought. 
For example, a slightly different value of weight decay can reduce the model robust accuracy by more than 7%, which is probable to override the potential promotion induced by the proposed methods. They conclude a baseline training setting and re-implement previous defenses to achieve new state-of-the-art results. These facts also appeal to more concerns on the overlooked confounders when benchmarking defenses. Code available on Github
#5 Towards Image Recognition with Extremely Low FLOPs with MicroNet
This paper presents MicroNet, an efficient convolutional neural network using extremely low computational cost (e.g. 6 MFLOPs on ImageNet classification).  MicroNet-M1 achieves 61.1% top-1 accuracy on ImageNet classification with 12 MFLOPs, outperforming MobileNetV3 by 11.3%.
It builds on two proposed operators: Micro-Factorized convolution and Dynamic Shift-Max. The former balances between the number of channels and input/output connectivity via low rank approximations on both pointwise and depth-wise convolutions. The latter fuses consecutive channel groups dynamically, enhancing both node connectivity and non-linearity to compensate for the depth reduction. Such a low cost network is highly desired on edge devices, yet usually suffers from a significant performance degradation. The work handles the extremely low FLOPs based upon two design principles: 
  • Avoiding the reduction of network width by lowering the node connectivity, and 
  • Compensating for the reduction of network depth by introducing more complex non-linearity per layer. 
Researchers hope it will provide good baselines for efficient CNNs on multiple vision tasks
Other Great AI Papers
A comprehensive evaluation framework that satisfies  performance specification of FL systems. Read more 
An efficient and scalable deep learning approach for road damage detection. Github Link  
A method for interpolating between generative models of the StyleGAN architecture in a resolution dependent manner. Generate Images from an Entirely Novel Domain
Building 3D Morphable Models from a Single Scan. New Research
Researchers propose open-world learning without labels. Towards Developing Autonomous True Open-world Never-ending Learning Agents
You should do these 5 things in your company before investing in AI. Read story 
Becoming a Machine Learning Engineer: Building  a Portfolio
Robotics in the Enterprise (free PDF) 
Best 9 Machine Learning Courses and Certifications. See here 
Gartner Top 10 Trends in Data and Analytics for 2020. Read more
Top AI News
The U.S. government needs to get involved in the A.I. race against China, Nasdaq executive says
Google Reveals Major Hidden Weakness In Machine Learning. Read more 
This AI model will help you summarize a research paper in seconds. Read more
AI Scholar Weekly
Happy New Month! If this newsletter lights you up, create a ripple effect by sharing it with someone else, and they will also be lit up! And, if you have suggestions, comments, or other thoughts, we would love to hear from you, email me at, tweet at @cdossman, like us on Facebook, or connect with me on LinkedIn
Did you enjoy this issue?
Educate AI

AI Scholar Weekly brings you everything new and exciting in the world of Artificial Intelligence, Machine Learning, and Deep Learning every week for free.

In order to unsubscribe, click here.
If you were forwarded this newsletter and you like it, you can subscribe here.
Powered by Revue