Discover more from Last Week in AI
Last Week in AI #166: New methods to detect deepfakes, AI for smarter farming, AI to detect heart problems from Apple Watch, and more!
New advances in deep fake detection by ByteDance and others, AI has many roles to play in improving farming efficiency, AI models can detect heart problems from a single ECG
Wang Weimin, a research scientist at ByteDance, won a five-month-long challenge to develop the best AI deepfake detector. The National University of Singapore graduate bested 469 other teams to win first place and a $100,000 cash prize in AI Singapore's Trusted Media Challenge--his detector could correctly identify deepfakes with 98.53% accuracy. Wang was offered another $300,000 as a start-up grant to commercialize his invention, but declined--Wang instead hopes to incorporate his AI model into the BytePlus platform as an offering to its clients.
Our Take: It's exciting to see continued efforts to combat deepfakes by way of competitions and novel methods. It's worth emphasizing that the deepfake problem is far from "solved." As more sophisticated methods for detecting deepfakes arise, so will more sophisticated methods for creating them. It is possible that some detection method is able to latch onto information that would be infeasible for a trickster to do anything about, but we aren't there yet. This is not to say that Wang's method won't be useful today--it likely will. In addition, Wang's choice to refuse the startup grant and instead use his AI model in BytePlus's platform is worth noting. State-of-the-art deepfake detectors seem like a public good, and the sort of outcomes I would like to see are either (a) an open-sourcing of the capabilities (not the internals) so that anyone can use them or, more likely, (b) an independent business that is willing to offer these detectors to anyone. I don't know precisely what the business model would look like and it is possible such a business could be spoofed by malicious actors looking to reverse-engineer the detection system (choice (a) also suffers from this issue), but deepfakes are a problem that plague many platforms and some sort of cooperation might help make progress.
Research into deepfake detection is also a very hot topic these days. See more recent advances here:
Solinftec’s AI platform called ALICE.AI allows customers to optimally schedule and plan their farming operations and make real-time farming decisions. ALICE.AI uses machine learning algorithms to provide insights to customers regarding equipment, weather conditions, and their crops all in real-time to help farmers understand and manage their crops better. Apart from raising $60 million in funding, they have also now announced a partnership with GROWMARK Inc for the development of a new robot to better monitor crop fields, as it is solar-powered and can run non-stop while sending data to ALICE.AI. They believe that autonomous scanning with the robot can increase yields, minimize wastage of agricultural input and also reduce environmental impacts. Solinftec’s technology currently manages more than 27 million acres of field crops like sugarcane, soy, corn, cotton, citrus, coffee, and timber in 11 countries around the world.
Our Take: With the global human population increasing, the projected food demand is to leap around 50% by 2050. Moreover, the world is experiencing climate changes that make resources like water and land scarcer. Using AI to improve agricultural efficiency and reduce environmental impact given the same amount of land is a promising direction, as its platform has cut over 680,000 tons of carbon dioxide emissions during the period 2016 to 2021.
Low ejection fraction — a heart condition that results in a reduced amount of blood pumped from the heart with each beat — affects 2~3% of people globally and is linked to worsening heart failures. This condition is usually difficult to detect as it rarely exhibits obvious symptoms, and hospitals traditionally needed an ECG with 12 electrodes to make a diagnosis. Mayo Clinic developed an AI algorithm that can identify this condition with just 1 electrode — an ECG on the Apple Watch. In a study across 11 countries over a period of 6 months, the company validated its algorithm and is now aiming to commercialize it for early disease detection.
Our take: It might’ve taken a few years but deep-learning-driven approaches in healthcare are now maturing and helping people in the real world. While the scope of such AI-healthcare applications may be limited — a single algorithm can probably only detect one type of disease and may give inaccurate predictions — in applications where the stakes are relatively low, like early disease detection, AI-powered algorithms have a real potential to make a difference.
Intel wins DARPA RACER-Sim program - "Intel expects these new simulation tools to significantly improve the development of autonomous systems using virtual testing, which reduces the risks, costs, and delays associated with traditional testing and verification protocols. "
In The Latest Research From The University Of Toronto, Machine learning-Based Approach Speeds Up The Counting Of Microplastics - "Researchers at the University of Toronto’s Faculty of Applied Science and Engineering have developed new techniques for counting and categorizing microplastics that employ machine learning to make the procedure quicker, faster, and cheaper."
Microsoft AI Researchers Develop MoLeR: A Deep Learning-Based Generative Model That Enables Efficient Drug Design - "Their paper demonstrates how generative models based on deep learning may aid in transforming the drug discovery process and uncovering new molecules more quickly."
Jumping robot leaps to record heights - "Research into biological and mechanical jumpers yields record breaking design"
Quantum-tunneling memory could boost AI energy efficiency by 100x - "There's a potential solution on the cards to the energy expenditure problems plaguing AI training, and it sounds simple: just strengthen the "synapses" that move electrons through a memory array."
Northeastern U & Microsoft Expand StyleGAN’s Latent Space to Surpass the SOTA on Real Face Semantic Editing - "Open-sourced by Nvidia three years ago, StyleGAN has wowed the Internet with its stunning human face synthesis capabilities. More recently, the powerful generative network has also shown its talents in semantic image editing, where it can modify a subject’s age, expression, gender, etc."
Tsinghua U & BAAI’s CogView2 Achieves SOTA Competitive Text-to-Image Generation With 10x Speedups - "Text-to-image generation has become one of the most publicly engaging AI research fields, with OpenAI’s recently unveiled state-of-the-art DALL-E-2 model garnering global mainstream media attention with its stunningly hyperrealistic images."
This Garage Project/ AI-Powered Note-Taking Tool Is Now Officially Microsoft Journal - "The researchers taught the app’s artificial intelligence (AI) to detect and categorize the things users type, such as headers, highlighted items, keywords, and even drawings. The software places a cue on the side of the screen for some of the drawings and titles, which users may tap to pick the material and then execute further actions like “move” or “copy.”"
Wearable, AI-powered whole-breast ultrasound system cleared by the FDA - "Though mammograms are seen as the current “gold standard” for breast cancer diagnosis, ultrasounds can actually provide a more accurate image of the breast tissue in younger women and others with d"
A unique machine-learning model predicts homelessness among US soldiers before their transition to civilian life - "Researchers led by Massachusetts General Hospital (MGH) and Harvard Medical School (HMS) have found that lifetime depression, trauma of having a loved one murdered, and post-traumatic stress disorder (PTSD) are the three greatest predictors of homelessness among U.S."
New machine learning maps the potentials of proteins - "An example of the shortest path between two proteins, considering the geometry of the graphing. By defining distances in this way, it is possible to achieve biologically more precise and robust conclusions. "
Nanomagnetic' computing can provide low-energy AI - "Researchers have shown it is possible to perform artificial intelligence using tiny nanomagnets that interact like neurons in the brain."
FDA plays tooth fairy, awards clearance to VideaHealth's cavity-spotting AI - "The tooth fairy isn’t the only one granting dentistry-related wishes."
Prototype Meta Headset Includes Custom Silicon for Photorealistic Avatars on Standalone - "Researchers at Meta Reality Labs have created a prototype VR headset with a custom-built accelerator chip specially designed to handle AI processing to make it possible to render the company’s photorealistic Codec Avatars on a standalone headset."
AI Model Can Accurately Detect Multiple Retinal Diseases in Real Time - "A new study shows that a deep-learning algorithm can recognize multiple retinal abnormalities simultaneously, which could help improve the early diagnosis of retinal diseases in underdeveloped areas."
Meet Nvidia's Huge H100 Chip That's Powering Tomorrow's AI - "What's happening Nvidia is releasing a new chip, the H100 "Hopper," that has the potential to speed up artificial intelligence that's sweeping the tech industry."
Ex-Googlers take a stab at building 'general intelligence' that makes software do what you tell it - "Adept AI, an artificial intelligence R&D lab founded by ex-Googlers who helped invent the popular transformer architecture, launched on Tuesday with the ambitious goal of teaching machines how to use "every software tool and API in the world."
SoundCloud acquires Musiio, an AI music curator, to improve discovery - "An online audio sharing platform and streamer, SoundCloud has acquired Musiio, an AI music curation company. With Musiio’s team and technology, SoundCloud plans to strengthen its music discovery features, helping the platform compete with streaming giants like Spotify."
Galileo exits stealth mode with $5.1M in funding to optimize AI training datasets - "Artificial intelligence startup Galileo Technologies Inc. today launched from stealth mode after raising a $5.1 million seed funding round led by The Factory. The Factory was joined by several angel investors including Anthony Goldbloom, the Chief Executive Officer of Google LLC’s Kaggle unit."
Another Firing Among Google’s A.I. Brain Trust, and More Discord - "The researchers are considered a key to the company’s future. But they have had a hard time shaking infighting and controversy over a variety of issues."
TAUGHT TO THE TEST - "AI software clears high hurdles on IQ tests but still makes dumb mistakes. Can better benchmarks help?"
UK Regulators' Path for AI Starts with Auditing Algorithms - "The Digital Regulation Cooperation Forum (DRCF), a group of four U.K. regulators, published last week, April 28 two documents where it provided businesses with guidance about the benefits and risks of artificial intelligence (AI) and machine learning (ML) and how to audit algorithms. "
Explainable AI: Language Models - "Just like a coin, explainability in AI has two faces — one it shows to the developers (who actually build the models) and the other to the users (the end customers). The former face (IE i.e. intrinsic explainability) is a technical indicator to the builder that explains the working of the model."
Researchers Taught A.I. to Write Wine and Beer Reviews and They're Pretty Convincing - "If you think that wine reviews seem to repeat the same words, relying on a familiar set of adjectives and other descriptors to convey what it's like to have a particular varietal in your glass, you're not entirely wrong."
T-SNE to view and order your Spotify tracks - "If like me you listen to a lot of different music and it's all mixed up together, it can be "surprising". Here's a little tool that allows you to view your music from Spotify and reorder it."
Copyright © 2022 Skynet Today, All rights reserved.