Categories
Machine Learning

Estimating the informativeness of data | MIT News

Researchers at MIT have developed a technique for estimating how much information data are likely to contain that is more accurate and scalable than earlier approaches. 

All data is not created equal. But how much information is likely to be included in any one piece of data? This issue lies at the heart of medical research, scientific experiment development, and even ordinary human knowledge and thought. Researchers at MIT have devised a novel method for addressing this issue, which has implications for health, scientific discovery, cognitive science, and artificial intelligence. The essential concept is to utilize probabilistic inference techniques to first infer which explanations are likely, and then use these probable explanations to generate high-quality entropy estimates rather than enumerating all possible explanations. The research demonstrates that this inference-based method is substantially quicker and more accurate than earlier methods. 

Source: https://news.mit.edu/2022/estimating-informativeness-data-0425 

Categories
Machine Learning

Meta AI Researchers Built An End-To-End Machine Learning Platform Called Looper, With Easy-To-Use APIs For Decision-Making And Feedback Collection | Mark Tech Post

Looper is a comprehensive artificial intelligence platform for optimization, personalization, and growth. 

Looper is a fully integrated artificial intelligence platform for optimization, customization, and feedback collecting developed by Meta Researchers. Looper can support the whole machine learning lifecycle, from model training through deployment and inference and product assessment and optimization. The Looper platform now hosts 700 AI models and generates 4 million AI outputs every second. It is intended for use cases with limited data volumes and model complexity that demand simplicity and speedy implementation. Looper runs in real-time, in contrast to many other AI systems, which make inferences in batch mode. 

Unlike large-scale AI models for vision, speech, and natural language processing, Looper uses models that can be rapidly retrained and deployed in huge numbers on shared infrastructure. A/B testing may investigate several models and decision rules, such as those employed by contextual bandits to mimic prediction uncertainty across one or more targets. A simple-to-use AI platform is often the deciding factor in adoption for teams with no previous expertise with production AI. Meta’s platform takes care of software updates, log management, and monitoring, resulting in considerable productivity improvements: It enables product makers to deploy more AI use cases than solutions with a restricted emphasis. Product teams have a spectrum of AI capabilities between beginners and expert AI engineers.

Source: https://www.marktechpost.com/2022/04/22/meta-ai-researchers-built-an-end-to-end-machine-learning-platform-called-looper-with-easy-to-use-apis-for-decision-making-and-feedback-collection/

Categories
Machine Learning

Machine Learning turns monochromatic night vision into a rainbow of colors | Daily Beast 

Thanks to machine learning, scientists are altering what we see when we gaze through a night vision scope. 

Researchers at the University of California, Irvine, employed machine learning to create a natural rainbow of hues from what you see via a night vision scope or camera. Not only may the research aid the military, but also medical technology, healthcare, and even specialized duties such as art restoration. To comprehend how the new night vision technology works, it’s necessary first to grasp human eyesight. Neural networks are computer programs that operate like that of an artificial brain. The researchers at UC Irvine predicted the visible spectrum picture using infrared photographs of three distinct wavelengths and deep learning. 

After that, the neural networks were tasked with reconstructing the pictures’ hue, which was now captured using a night vision camera. Artificial neural networks will enable a slew of various scientific application undertakings. While the military is undoubtedly interested in this technology, it might also be beneficial in eye surgery and art restoration. “They have the potential to improve a clinician’s capacity to operate,” Browne said. “When applied to new technologies, they can improve the technology’s performance.” 

Source: https://www.thedailybeast.com/machine-learning-turns-monochromatic-night-vision-into-a-rainbow-of-colors  

Categories
Machine Learning

AI confirms the obvious: The pandemic bummed people out | Popular Science

Scientists created a mood ring for the internet using machine learning, and it revealed how depressed the world was during the outset of the epidemic. 

Researchers use mood as a unique tool to assess the effect of natural and man-made catastrophes on humans. However, in the aftermath of a major incident, it’s just impracticable to ask every single individual on the planet how they’re feeling. Scientists from MIT, the Chinese Academy of Sciences, and the Max Planck Institute for Human Development devised a solution. They employed machine learning methods to monitor social media for mood changes after the initial wave of COVID-19 in 100 different nations, allowing them to gather real-time readings on how pleased or upset the pandemic’s occurrences made individuals all across the globe. Consider it as an AI-powered mood ring, but for millions of people. Their results were published in the journal Nature Human Behavior only last week. 

Source: https://www.popsci.com/technology/machine-learning-measures-social-media-sentiment-mit/ 

Categories
Machine Learning

A New Type of Hand Prosthesis Learns From the User, and the User Learns From the Prosthesis | Neuroscience News

Bionic Arm Interfaces Become More Robust Due to New Technology

Dennis Yeung, a Ph.D. student at Aalto University, and his research group have invented a new form of technology that improves the compatibility of a prosthesis with the severed region. Individuals with an amputated upper limb may operate the robotic prosthesis by contracting their residual muscles. Advanced prostheses include machine learning techniques to assist in interpreting these user-generated signals. These connections are often susceptible to external causes, such as sweat, and deteriorate with time. Unsupervised adaptive myocontrol (UAM) is a myoelectric interface that extracts natural motor synergies from multi-muscle data and adjusts to new user inputs in real-time.

UAM has been evaluated by a series of virtual target reaching activities done by healthy and amputee volunteers. Tests were done under normative and electrode perturbed settings to assess control robustness.

Source: https://neurosciencenews.com/learning-arm-prosthetics-20225/