AI is one of the most happening domains in the world right now. It would take a lifetime to skim through all the machine learning research papers released till date. As the AI keeps itself in the news through new releases of frameworks, regulations and breakthroughs, we can only hope to get the best of the lot.
So, here we have a compiled a list of top exciting AI announcements released over the past one week:
Google Launches TensorFlow For Quantum Computing
Late last year, Google locked horns with IBM in their race for quantum supremacy. Though the news has been around how good their quantum computers are, not much has been said about the implementation. Today, Google brings two of their most powerful frameworks Tensorflow and CIRQ together and releases TensorFlow Quantum, an open-source library for the rapid prototyping of quantum ML models.
TFQ is designed to provide the developers with the tools necessary for assisting the quantum computing and machine learning research communities to control and model quantum systems.
A key feature of TensorFlow Quantum is the ability to simultaneously train and execute many quantum circuits. This is achieved by TensorFlow’s ability to parallelise computation across a cluster of computers, and the ability to simulate relatively large quantum circuits on multi-core computers.
DeepMind Uses Its AlphaFold System
As the devastating news of COVID-19 keeps rising at an alarming rate, the AI researchers have given something to smile about. DeepMind, one of the premier AI research labs in the world, announced last week, that they are releasing structure predictions of several proteins that can promote research into the ongoing research around COVID-19. They have used the latest version of AlphaFold system to find these structures. AlphaFold is one of the biggest innovations to have come from the labs of DeepMind, and after a couple of years, it is exhilarating to see its application in something very critical.
Google Releases Most Efficient NLP Model
As the pursuit to achieve human-level intelligence in machines fortifies, language modeling will keep on surfacing till the very end. One, human language is innately sophisticated, and two, training language models from scratch is exhaustive.
The last couple of years has witnessed a flurry of mega releases from the likes of NVIDIA, Microsoft and especially Google. As BERT topped the charts through many of its variants, Google now announces ELECTRA.
ELECTRA has the benefits of BERT but more efficient learning. They also claim that this novel pre-training method outperforms existing techniques given the same compute budget.
The gains are particularly strong for small models; for example, a model trained on one GPU for four days outperformed GPT (trained using 30x more compute) on the GLUE natural language understanding benchmark.
Alibaba’s AI Detects With 96% Accuracy
China has been the worst-hit nation of all the COVID-19 victims. However, two of the biggest AI breakthroughs have come from the Chinese soil. Last month, Baidu announced how its toolkit brings down the prediction time. Last week, another Chinese giant, Alibaba announced that its new AI system has an accuracy of 96% in detecting the coronavirus from the CT scan of the patients. Alibaba’s founder Jack Ma has fueled the vaccine development efforts of his team with a $2.15 M donation.
Convert Any 2D Photo Into 3D Using CNN
Facebook AI has released its in-house feature of converting a two-dimensional photo into a video byte that gives the feel of having a more realistic view of the object in the picture. This system infers the 3D structure of any image, whether it is a new shot just taken on an Android or iOS device with a standard single camera, or a decades-old image recently uploaded to a phone or laptop.
The feature has been only available on high-end phones through the dual-lens “portrait mode”. But, now it will be available on every mobile device even with a single, rear-facing camera. To bring this new visual format to more people, the researchers at Facebook used state-of-the-art ML techniques to produce 3D photos from virtually any standard 2D picture.
One significant implication of this feature can be an improved understanding of 3D scenes that can help robots navigate and interact with the physical world.
Honeywell Claims It Has The Best Quantum Computer
As the whole world focused on the race to quantum supremacy between Google and IBM, Honeywell silently has been building, as it claims, the most powerful quantum computer yet. And, it plans to release this by the middle of 2020.
“Thanks to a breakthrough in technology, we’re on track to release a quantum computer with a quantum volume of at least 64, twice that of the next alternative in the industry. There are a number of industries that will be profoundly impacted by the advancement and ultimate application of at-scale quantum computing,” said Tony Uttley, President of Honeywell Quantum Solutions in the official press release.
Corona Effect Hits Conferences
The outbreak of COVID-19 has created a panic globally and rightfully so. Many flagship conferences have been either cancelled or have been moved to a virtual environment.
Nvidia’s flagship GPU Technology Conference (GTC), which was supposed to take place in San Francisco in the last week of March was cancelled due to fears of the COVID-19 coronavirus.
Whereas, Google Cloud also has cancelled its upcoming event, Google Cloud Next ’20, which was slated to take place on April 6-8 at the Moscone Center in San Francisco. “Due to the growing concern around the coronavirus (COVID-19), and in alignment with the best practices laid out by the CDC, WHO and other relevant entities, Google Cloud has decided to reimagine Google Cloud Next ’20,” the company stated on its website.
One of the popular conferences for ML researchers, ICLR2020 too, has announced that they are cancelling its physical conference this year due to growing concerns about COVID-19. They are shifting this event to a fully virtual conference.
ICLR authorities also issued a statement saying that all accepted papers at the virtual conference will be presented using a pre-recorded video.