Technology is developing today at a pace that's never been seen before. New advancements and breakthroughs happen far more readily than at any time in the past. One of the most talked-about areas of cutting-edge tech is that of artificial intelligence (AI). AI is driving the digital transformation of organizations in all manner of niches. So wide-ranging are the applications of AI, that you've probably already interacted with an example of the tech today.

Amazon SageMaker is a fully managed service that provides every developer and data scientist with the ability to build, train, and deploy machine learning (ML) models quickly. Amazon SageMaker removes the heavy lifting from each step of the ML process to make it easier to develop high-quality models. As datasets continue to increase in size, additional compute is required to reduce the amount of time it takes to train. One method to scale horizontally and add these additional resources on Amazon SageMaker is through the use of Horovod and Apache MXNet. In this post, we show how you can reduce training time with MXNet and Horovod on Amazon SageMaker.

Cyber threats continue to escalate in both sophistication and volume. Traditional approaches to threat detection, however, are no longer sufficient to ensure protection. Correspondingly, machine learning (ML) has proven highly effective at identifying and warding off cyber attacks. Machine learning's power is the result of three factors: data, compute power and algorithms. Due to its very nature, the cyber field produces substantial amounts of data.

The global pandemic has been cited as a "wake-up call" for many things -- the environment, economic and social rights, and general global inequalities. However, scientist, author, and entrepreneur Gary Marcus thinks that the COVID-19 crisis should also be considered a wake-up call for AI. Speaking at the virtual Intelligent Health AI conference yesterday, Marcus lamented decades of missed opportunities to build a more robust artificial intelligence, arguing that too much attention has been placed on AI technologies that don't really help the world in any meaningful way. "We would like AI that could read and synthesize the vast, quickly growing medical literature, for example, about COVID-19," he said. "We want our AI to be able to reason causally, we want it to be able to weed out misinformation. We want to be able to guide robots to keep humans out of dangerous situations, care for the elderly, deliver packages to the door. With AI having been around [for] 60 years, I don't think it's unreasonable to wish that we might have had some of these things by now. But the AI that we actually have, like playing games, transcribing syllables, and vacuuming floors, it's really pretty far away from the things that we've been promised."

The leading pre-trained language models demonstrate remarkable performance on different NLP tasks, making them a much-welcomed tool for a number of applications, including sentiment analysis, chatbots, text summarization, and so on. However, good performance usually comes at the cost of enormous computational resources that are not accessible by most researchers and business practitioners. To address this issue, different research groups are working on increasing the compute-efficiency and parameter-efficiency of the pre-trained language models without sacrificing their accuracy. Among the novel approaches introduced this year, at least three methods are appraised by the AI community as very promising. To help you stay aware of the latest NLP research advancements, we have summarized the corresponding research papers in an easy-to-read bullet-point format.

We asked GPT-3, OpenAI’s powerful new language generator, to write an essay for us from scratch. The assignment? To convince us robots come in peace

Last week, The Guardian ran an op-ed that was supposedly written by GPT-3. The article was very misleading and had zero value in informing the public about advances in AI. But it perfectly showed us how humans and AI can team up to create sensational and moneymaking BS. Here's why I think the entire methodology was very wrong, misleading and damaging to AI research: In case you want to read the original article on Guardian (IMO don't waste your time):

The growth of AI/Deep learning and data analytics has created many of the most challenging HPC workloads in recent years. The latest HPC report by Hyperion Research states that iterative simulation workloads and new workloads such as AI and other Big Data jobs would drive the adoption of HPC storage. To keep up with the growing massive amount of data we are collecting, users need to enhance computation performance at the same time and hence HPC requires equally robust storage to maintain compute performance for faster data in and out as we are heading into the Big Data era now. Data-intensive HPC is driving new storage requirements and making a change. For the simulation process, it not only requires a large amount of computations running on HPC infrastructures built on a cluster of powerful servers linked together with networking and memory, but also adds in self-service concept data stores.

Box filters computed using integral images have been part of the computer vision toolset for a long time. Here, we show that a convolutional layer that computes box filter responses in a sliding manner can be used within deep architectures, whereas the dimensions and the offsets of the sliding boxes in such a layer can be learned as part of an end-to-end loss minimization... Crucially, the training process can make the size of the boxes in such a layer arbitrarily large without incurring extra computational cost and without the need to increase the number of learnable parameters. Due to its ability to integrate information over large boxes, the new layer facilitates long-range propagation of information and leads to the efficient increase of the receptive fields of downstream units in the network. By incorporating the new layer into existing architectures for semantic segmentation, we are able to achieve both the increase in segmentation accuracy as well as the decrease in the computational cost and the number of learnable parameters.

The SNN approach uses biologically inspired, event-driven spike-based computation and communication -- meaning only operating when needed -- in its design. One of the distinguishing features of SNN as a computing paradigm is the integration of the element of time into algorithms and models. Penn State scientists are exploring novel magnetic device structures to directly mimic such temporal non-linear characteristics in hardware, scalable architecture and interconnection fabrics for these devices, along with novel hybrid algorithm designs to leverage the benefits of both SNN models and traditional non-spiking deep learning models.