Usually, every month we write an article about the best and most promising AI research papers from that month. You can check them out here:
In addition to that, we list fifteen AI articles we have found amazing that month. This collection of articles should give you an overview of what happened that month in the AI industry both from technical, business and from an ethical perspective. You can check out the April edition here. Have fun!
Not so long ago, Andrej Karpathy famously tweeted: “Gradient descent can write code better than you. I’m sorry.” What he was trying to say is that neural networks, which use Gradient descent optimization technique, will soon be able not just to write code, but to write code better than us – software developers. Stay relevant in the rising AI industry an learn all you need to know about deep learning here!
Watson’s Creator Wants to Teach AI a New Trick: Common Sense
David Ferrucci, the man who built IBM’s Jeopardy!-playing machine, Watson, is explaining a children’s story to his new creation. His asks seemingly simple yet hard question “Can we ever get machines to actually understand what they read?”. It is the question that starts most of the AGI discussions.
Announcing Meta-Dataset: A Dataset of Datasets for Few-Shot Learning
A few-shot classification is the classification method in which the training set contains classes that are completely different from those that will be available at test time. In this particular case, the training process builds a model that can classify new classes with only several examples. With Meta-Dataset, a new dataset created by Google, you can create this kind of models.
AI advances to better detect hate speech
Guys from Facebook AI released this article in which they explain how they fight against hate speech on their network. This is one of the complex and important tasks that they need to do. For this, they use various AI techniques that are listed in this article. Now, these techniques proactively detect 88.8 percent of the hate speech content that they remove.
AI systems are worse at diagnosing disease when training data is skewed by sex
Removing bias from data is one of the most important things we should consider when we are building AI systems. This is especially the case if we create it for some kind of healthcare system. In this article, the authors discuss the dangers of not having diverse data in your dataset and the effects of such a mistake.
The Big Bad NLP Database
The title says it all. This is a site that contains a huge number of NLP datasets. It is like a playground for engineers who are building this type of system. Have fun peeps, have fun!
40 Years on, PAC-MAN Recreated with AI by NVIDIA Researchers
Do you like Pac-Man? What a stupid question, of course you like Pac-Man. Engineers from NVIDIA liked it so much that they rebuild it, but only using the AI. They used their Generative Adversarial network GameGAN that was trained on 50,000 episodes of the game. Beyond cool!
Artificial Intelligence Consumes a Startling Amount of Power – MIT System Reduces the Carbon Footprint
Training AI systems require a lot of processing power. A lot of processing power equals a lot of energy spent, which in turn is bad for the environment. This is why cool kids from MIT developed a new automated AI system with improved computational efficiency and a much smaller carbon footprint.
Old tools, new tricks: Improving the computational notebook experience for data scientists
In this article, a team from Microsoft discusses problems of Computational notebooks and gives suggestions on how to improve. The problems that are discussed are gathered from semi-structured interviews, followed by a survey with over 150 data scientists.
This awesome project is creating cartoon “deep fakes”, ie. 2D vector illustration and animates its containing curves in real-time based on the recognition result from PoseNet and FaceMesh. It borrows the idea of skeleton-based animation from computer graphics and applies it to vector characters.
TensorFlow 2.2.0 Released
TensorFlow 2.2.0 is released four months after 2.1.0. In this article, you can find the list of all features and changes that are available in the new subversion.
OpenAI debuts gigantic GPT-3 language model with 175 billion parameters
Yes, it is a 175 billion parameter, 700Gb NLP model coming from OpenAI. It achieves awesome results even when it is not pre-trained.
Facebook AI, AWS partner to release new PyTorch libraries
Interesting news from Facebook and AWS. They have partnered to develop new libraries targeted at large-scale elastic and fault-tolerant model training and high-performance PyTorch model deployment. We can’t wait to see the results. How about you?
COVID-19 Research Explorer
Google‘s AI team has released a new tool to help researchers traverse through a trove of coronavirus papers, journals, and articles. The COVID-19 research explorer tool is a semantic search interface that sits on top of the COVID-19 Open Research Dataset (CORD-19).
TFRT: A new TensorFlow runtime
Probably the biggest complaint about TensorFlow is its speed. That is why along with 2.2.0 Google presented TFRT – a new runtime that will replace the existing TensorFlow runtime. It is responsible for the efficient execution of kernels on targeted hardware.
About positive psychology and ethics in AI
Building Smart Apps