News by Dmytro Spodarets
467 items
Video-based advance learning to play Minecraft (VPT)
Video Pre-Play Training (VPT) is a new semi-supervised simulation learning method that helps you leverage the wealth of unlabeled video data available on the Internet for learning. Based on this data, an IDM backward dynamics model is trained, which can be used to label vast amounts of online video data and learn how to act by behavioral cloning. The method is tested on Minecraft, which is one of the most actively played video games in the world. A trained behavioral cloning model ("basic VPT")
Jul 6, 2022
Rules for the use of artificial intelligence technology
The European Union is working on amendments to the Artificial Intelligence Act - it will regulate the use of AI detectors and ban polygraph systems at borders. This ban is just one of thousands of amendments to the Artificial Intelligence Act being considered by the EU officials and members of the European Parliament. Among those influencing changes to the Act are human rights activists, labor unions, and Google and Microsoft. They all want the AI Law to distinguish between creators of general-
Jul 4, 2022
Nvidia announced the transformation of data centers into AI factories
Nvidia reported a late series of hardware milestones and modernization to support advancing the company’s aspirations. The most important technology that will help implement Nvidia’s vision is the company’s Grace superchips. At Computex, Nvidia reported that various hardware vendors, including ASUS, Foxconn Industrial Internet, GIGABYTE, QCT, Supermicro and Wiwynn, will build Grace base systems that are supposed to start shipping in the first half of 2023. Grace superchip should be available i
Jun 3, 2022
Improvements of OpenAI with a New DALL-E Model
DALL-E 2 is a new model that can take complex enough prompts. It should comply and search for outputs which are likely to meet the user’s requirements out of multiple existing outputs. The most exciting part of this model is its leap towards a diffusion model, a type of image production that in the beginning looks like pure noise and changes the image over time, repeatedly creating it a little more like the image requested until there’s no noise left at all. OpenAI takes into account the risk
Jun 2, 2022
When Apple Made Him Go to the Office He Went to Google
Ian Goodfellow, one and only one who oversaw machine learning and artificial intelligence at Apple, simply left the company a few weeks ago, because of the lack of flexibility in its office policies. Goodfellow’s decision to start working for Google is a coup for the DeepMind division, which puts him on as an individual contributor. Ian is well known as one of the first machine learning researchers, and the move is a reunion of sorts. As of today, Goodfellow is the most senior employee known t
Jun 1, 2022
New Implementation of PyTorch Training for Mac
After a new PyTorch v1.12 release, creators and researchers can now benefit from Apple’s silicon GPUs to enable an essentially faster model training. They gain the ability to implement machine learning workflows like modeling and fine-tuning locally, right on their Macs. The MPS backend develops the PyTorch framework, providing scripts and capabilities to arrange and run operations on Mac. MPS improves compute performance with kernels that are fine-tuned for the unique characteristics of every
May 30, 2022
Resolving Math Olympiad Tasks
The program called statement curriculum learning [https://openai.com/blog/formal-math/] features the capabilities that enable manual collection of sets of declarations of different difficulty levels, where the most difficult statements are identical to the benchmark of their target. However, as demonstrated, a neural prover itself is fragile and can only prove a few of these statements. The company addresses the problem of a limitless action space obstacle by sampling actions from a language mo
Feb 23, 2022
AI & ML Collaborate with Healthcare Industry
A new research by CSA found that the advance of AI, ML, and data mining in healthcare [https://venturebeat.com/2022/01/18/report-how-ai-and-ml-optimize-diagnosis-process-in-health-care/] has considerably accelerated. The companies invent to build intelligent systems, to upgrade the diagnosis process and capture unexpected patterns within complex and large datasets. By using AI and ML, healthcare service providers can achieve the level of precision diagnostics they are looking for. The medical
Feb 21, 2022
DeepMind Develops AlphaCode for Competitive Programming
DeepMind has announced that it developed a novel system called AlphaCode that can compose computer programs at a competitive level [https://deepmind.com/blog/article/Competitive-programming-with-AlphaCode]. AlphaCode accomplished a predicted rank within the top 54% of candidates in programming competitions by fixing new problems that require a merge of critical thinking, logic, algorithms, coding, and natural language comprehension. The team authorized their appearance by using competitions pla
Feb 19, 2022
EleutherAI’s Announces GPT-NeoX-20B
EleutherAI’s has announced their newest open-source language model: GPT-NeoX-20B [https://blog.eleuther.ai/announcing-20b/]. It is a 20 billion parameter model trained by using the GPT-NeoX framework on GPUs generously provided by the team from CoreWeave. GPT-NeoX-20B is the biggest candidly available educated general-purpose autoregressive language model, and the team predicts it to work great on many tasks. In addition to that, it is going to be created a 20b channel in Discord for conversati
Feb 18, 2022
New Medications Help Avoid COVID-19
Scientists have successfully created a new kind of drugs, helping to prevent COVID-19. The drug stimulates the creation of proteins that can turn on and off specific genes that fight the disease. The proteins contain hundreds to thousands of smaller building blocks called amino acids. Amino acids itself are connected to one another in long chains that fold up to form a protein. The scientists managed to create these proteins artificially. The method they used is based on the mechanism that pass
Feb 12, 2022
New Family Member but It’s a Digital Assistant
According to research, any device can be more proficient and emotionally alluring to families if it is capable of exhibiting various social cues, such as looking at the speaker during conversation. Moreover, the researchers have discovered that a marketing part, especially the manufacturer’s name, which is associated with the device, has a considerable effect on how families comprehend and work with voice-user interfaces. The experts want to keep exploring how families co-use voice-user program
Feb 11, 2022