AI to Hunt Down Bioactive Peptides in Food

Nuritas, a biotechnology company, wants to capitalize on discovery of natural active ingredients that can improve human health. The company uses AI to find out peptides present in the food that we eat. Using AI for this increases the speed and accuracy of peptide identification. High speed identification is critical as it avoid the development of flawed candidates. Peptides provide extraordinary benefits to human health. The whole project is segmented into 3 stages. First step defines the health condition being targeted to improve chances of finding an effective peptide. In second step, AI is used to predict the food-derived bioactive peptides. Finally, the third step is to unlock the found peptide from its food source so that it can be used therapeutically.


A Quantum Algorithm That Could Make AI Better

According to a research at National University of Singapore (NUS), ‘Artificial Intelligence’ and ‘Quantum Computing’ are two recent phenomena with massive potential, but both are at its very nascent stage and aren’t as perfect yet. As per researchers from the Center of Quantum Technologies at NUS, AI’s current form limited to specialized machine learning algorithms can be greatly improved by quantum computing. They propose a quantum linear system algorithm which will allow faster analysis of larger data sets through quantum computers. Like traditional algorithms, quantum algorithms are a step-by-step procedure; however, they use features specific to quantum computing, such as quantum entanglement and superposition. AI systems and their algorithms would get a boost from quantum computing through ‘quantum information processing’, speeding up classical ML tasks.

 


 

AI Succumbing to Racial Discrimination Based on Gender and Skin Colour

According to a new researcher from MIT and Stanford University, AI powered facial-analysis programs from major technology companies suffers from both skin-type and gender biases. Error rates in determining the gender of light-skinned men had upper limit of 0.8% while the same for darker-skinned women climbed to 20% and 34% for two different experiments. This puts neural network technology in question, especially if it is used to detect gender involved in a criminal activity. While companies claim success rate of 97%, upon further investigation, it is revealed that the dataset used was highly skewed with more than 77% males, of which about 83% were white males amongst them.


MIT Develops a New Chip that Speeds Up Neural-Network Computations

Most recent advances in AI is due to neural network technology. But they have high energy consumptions. Researchers at MIT have developed a chip that increases the speed of neural-network computations by three to seven folds. This results in approximately 95% reduction in power consumption. Conventionally, data is moved back and forth between memory and processor of the chip. With calculations running into thousands, this transfer of data consumes lion’s share of energy. Researchers have now proposed to use new specific operation, called the “dot product functionality”, which eliminated the need of data transferring. Avishek Biswas, an MIT graduate student along with Anantha Chandrakasan, dean of MIT’s School of Engineering is scheduled to present a research paper on this at International Solid State Circuits Conference.

 


 

Machine Learning Tool Predicts Output of Organic Reactions

Researchers at IBM have developed an algorithm that can predict end products of organic chemistry reactions. The algorithm is based on AI’s neural network technology. The algorithm had correctly predicted the outcome of the reaction with accuracy of 80%. This was reached without giving any training of organic chemistry rules to it. Instead, the team gave it more than 50,000 patented reactions to learn from. Researcher Philippe Schwaller explains, “From the reactant plus the reagents, it tries to guess the most likely product. By showing it the same training set again and again, it slowly learns how to construct a valid product.” Computer first converts the chemical structure of a reactant into a string of letters and then treats the reaction like a translation problem. It uses the algorithm originally developed for language processing.


ML Algorithms for Microblogging Sites to Weed Out ISIS Propagandas

US government has allocated special funding for creating machine learning algorithm to detect ISIS propaganda videos online. London-based startup, ASI Data Science has created a tool and its costs $830,000. Smaller video platform like Vimeo and pCloud are currently the target websites. Algorithm can detect 94 percent of ISIS propaganda with 99.99 percent accuracy, claims the company. Factors taken into the algorithm aren’t made public by them till now. But the most probable explanation is that it might include visual cues, like logos, but also metadata, like where the video was uploaded from.

 


 

How Individuals By-pass Detection Algorithms on Social Network?

Dr. Talal Rahwan, Assistant Professor-Computing Science at Khalifa University, in his research paper has claimed that while social media giants have tried to tackle the problem of promoting extremism through their platforms, the system is still vulnerable. The paper expressed how criminals and terrorist groups dodge their detections by creating fake relationships that dupe network-analysis tools. He says, “We were investigating how people manage their connections to evade such network-analysis tools.” He demonstrates the fragility of ‘node centrality’ analysis which is deployed by security agencies to flag leader in a social network. The finding highlights that around 55% of people accept friendship requests from strangers. Fake connections go a long way in escaping analysis tools set by authorities to find out such extremists, thus leaving it exposed.


AI Powered Brain Scanning to Assist in Treatment Of OCD

Obsessive compulsive disorder (OCD), is a lifelong illness marked by repetitive thoughts and actions. Conventional treatment includes psychotherapy and medication. But the success rate isn’t very promising. UCLA researchers have developed a way to use brain scans along with machine learning algorithms to predict if conventional cognitive behavior benefits people with OCD. Using a functional MRI, researcher scanned brains of 42 people with OCD, before and after four weeks of cognitive behavioral therapy to determine how different areas of the brain activate in sync with each other. Alongside, they assessed severity of symptoms using a scaled system where a lower score indicates less severe or less frequent symptoms. The algorithm predicted which patients would fail to respond to cognitive behavioral therapy with 70% accuracy.

 


 

ML to Make Online Dating A Pleasant Experience

With massive surge in online scammers on online dating platform, there is no respite for members seeking out for their companion. As per survey, in the past three years in America alone, online dating scam costs have gone up to $20M. According to Kevin Lee, architect at fraud prevention and machine learning firm Sift Science, an online scam starts with a genuine conversation at first and then reach a stage termed as an ‘extraction point’ where the scammer asks the other party for money to 'help them out.' He is working on a solution where combination of analytics and ML identifies users who have multiple profiles by tracing the account specific IP addresses, devices, etc. Content analysis is done as well, where texts suggesting requests for money are flagged out as a potential scam.


Disclaimer

Click Here