View profile

AI Scholar Weekly - Issue #54

AI Scholar Weekly - Issue #54
By Educate AI  • Issue #54 • View online
Facebook’s New Billion-Parameter Model Might Just Change Computer Vision Forever; Deep Learning was Top In-demand Skill of 2020; A Crash Course on Reinforcement Learning; Deep Learning in Medical Imaging (and more)

Top AI Research Papers This Week!
#1 A New Multi-agent Approach for Adversarial Environment Generation
In collaboration with the University of California Berkeley AI Research, Google AI has proposed a new multi-agent approach for training the adversary, a publication recently presented at NeurIPS 2020.
The study presents an algorithm, Protagonist Antagonist Induced Regret Environment Design (PAIRED), that’s based on minimax regret and prevents the adversary from creating impossible environments while still enabling it to correct weaknesses in the agent’s policy.
PAIRED incentivizes the adversary to tune the generated environments’ difficulty to be just outside the agent’s current abilities, leading to an automatic curriculum of increasingly challenging training tasks. 
According to the researchers, agents trained with PAIRED learn more complex behavior and generalize better to unknown test tasks. 
Results: show its relevance to a range of RL tasks, from learning increasingly complex behavior or more robust policies to improving generalization to novel environments.
They have released open-source code for PAIRED on our GitHub repo. Read paper; Emergent Complexity and Zero-shot Transfer via Unsupervised Environment Design.
#2 Fast Interactive Video Object Segmentation with Graph Neural Networks
In this paper, researchers show that a graph neural network-based approach can achieve state-of-the-art results in interactive video object segmentation with a significantly smaller number of trained parameters and less training data than most existing solutions. 
Interactive video object segmentation aims to utilize automatic methods to speed up the process and reduce the annotators’ workload. 
The proposed network operates on super pixel-graphs, allowing users to reduce the problem’s dimensionality by several magnitudes. Researchers show that the network possessing only a few thousand parameters can achieve state-of-the-art performance. At the same time, inference remains fast and can be trained quickly with very little data without overfitting. Additionally, the method is fast, both in terms of training and inference. The software code and the trained model can be accessed here
#3 Deep Learning in Medical Imaging: Traits, Trends, Case studies Highlights, and Future Promises
Deep learning has been widely used in various medical imaging tasks. It has achieved remarkable success in many medical imaging applications, thereby boosting us into the so-called artificial intelligence era.
This research presents medical imaging traits, highlights both clinical needs and technical challenges in medical imaging, and describes how emerging trends in deep learning are addressing these issues. 
Researchers cover network architecture topics, sparse and noisy labels, federating learning, interpretability, uncertainty quantification, etc. They also present several case studies applications commonly found in clinical practice, including digital pathology and chest, brain, cardiovascular, and abdominal imaging, and conclude with a discussion and presentation of promising future directions. Read more: A Survey of Deep Learning in Medical Imaging
#4 Unsupervised 3D shape reconstruction from 2D Image GANs
Natural images are projections of 3D objects on a 2D image plane. While state-of-the-art 2D generative models like GANs show unprecedented quality in modeling the natural image manifold, it is unclear whether they implicitly capture the underlying 3D object structures. And if so, how could we exploit such knowledge to recover the 3D shapes of objects in the images? 
To answer these questions, this paper presents the first attempt to directly mine 3D geometric indications from an off-the-shelf 2D GAN trained on RGB images only. 
We found that such a pre-trained GAN indeed contains rich 3D knowledge and thus can be used to recover 3D shape from a single 2D image in an unsupervised manner, the researchers say 
The framework’s core is an iterative strategy that explores and exploits diverse viewpoints and lighting variations in the GAN image manifold. It does not require 2D keypoint or 3D annotations or strong assumptions on object shapes, yet it successfully recovers 3D shapes with high precision for human faces, cats, cars, and buildings. The recovered 3D shapes immediately allow high-quality image editing like relighting and object rotation. 
Results; demonstrate the effectiveness of the proposed approach compared to previous methods in both 3D shape reconstruction and face rotation. Get the GAN2Shape code here
#5 A Deep Emulator for Secondary Motion of 3D Characters
Fast and light-weight methods for animating 3D characters are desirable in various applications such as computer games.
Researchers present a learning-based approach to enhance skinning-based animations of 3D characters with vivid secondary motion effects. They have designed a neural network that encodes each local patch of a character simulation mesh where the edges implicitly encode the internal forces between the neighboring vertices.
The network emulates the ordinary differential equations of the character dynamics, predicting new vertex positions from the current accelerations, velocities, and positions. Being a local method, the proposed network is independent of the mesh topology and generalizes to arbitrarily shaped 3D character meshes at test time.
Results: On evaluation, this method can be over 30 times more efficient than ground-truth physically-based simulation, and outperforms alternative solutions that provide fast approximations.
Other Great AI Papers
We have witnessed an increase in cyber threats and malicious software attacks on different platforms with important consequences to persons and businesses. This paper proposes the use of a Transformers’ architecture to automatically detect malicious software. 
A new model watermarking framework for protecting deep networks trained for low-level computer vision or image processing tasks. Read full paper
A rapport-building 3D embodied conversational agent for the Human Support Robot. Read more
A comparative review of VAEs, GANs, normalizing flows, energy-based and autoregressive models. These techniques are drawn under a single cohesive framework while reviewing current state-of-the-art advances and implementations
This is the first systematic and comprehensive review of automated machine learning on graphs. This paper shares our insights on future research directions for automated graph machine learning. 
AI Resources
This is a self-contained repository to explain two basic Reinforcement (RL) algorithms. A Crash Course on Reinforcement Learning
Highly Recommended AI and ML Books for Business Leaders. Go here
The top 3 Effective Feature Selection Strategies. Click to read
Top AI News
OpenAI’s state-of-the-art machine vision AI is fooled by handwritten notes. Read story
Deep Learning was Top In-demand Skill of 2020, List of Most Popular Nanodegree Program. Read more
Facebook’s New Billion-Parameter Model Might Just Change Computer Vision Forever. Read more
AI Can Now Learn What Faces You Find Attractive Directly From Your Brain Waves. Read more
AI Scholar Weekly
Thanks for reading. Create a ripple effect by sharing this AI Scholar Newsletter with someone else, and they will also be lit up!
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