Singularity Pulse  
     
  Finance

AI Enabled Trading Bot

JP Morgan asset management team is working on developing a new software to execute profitable trades with more efficiency, compared to human traders. Lee Bray, head of equity trading says,” By creating a systematic, adaptive model able to alter actions based on mathematical patterns rather than relying on human input, we’re transitioning equity trading to be more scientific and quantifiable.” In its current form, the algorithm can offer recommendations which are used by human traders. It is also able to perform other administrative work, providing traders with more efficiency.

 
  Image  
 

Google AI Can Isolate A Single Voice from Crowd

Researchers at Google have been working on an algorithm to isolate speech in videos. Method of training the algorithm consisted of building fake “cocktail parties” consisting manually spliced "clean" sources of audio and video, overlaid with background noise. Data is then fed to the network to train it with facial movements. Algorithm is able to determine which frequencies are most likely corresponding to a given speaker. Data is then extracted to create a new isolated audio track, free from noises.

Information Technology  
  Image  
  Information Technology

AI to Detect ‘Deepfake’ Face Swaps Online

Researchers from Germany’s Technical University of Munich have developed an algorithm called XceptionNet that quickly spots faked videos posted online. The team started training their deep-learning neural network with a dataset of more than 1,000 videos and 500,000 images. ML was able to figure out the differences between the doctored and un-doctored images. The convolution neural network was able to correctly identify fake videos with accuracy from 87% to 98%.

 
  Image  
 

AI to Drive “Mjunction” Trading

A joint venture between Tata Steel and SAIL has developed a ML powered platform “Mjunction” which will be used for electronic auction of e-commerce firms. Mjunction handles around 150 auctions per day. The algorithm is aimed at improving auction process and provide better price discovery mechanism. The software will not only find right bidders for auctions, but also relevant auctions for bidder. Data gathered through previous bidding history, location of the bidder, size of auction lots, number of buyers etc. are used as inputs to make predictions.

Manufacturing  
  Image  
  Healthcare

AI Device for Diabetic Eye Problems

In a first of its kind, US regulators have approved a device which uses artificial intelligence to detect eye damage from diabetes. The device is called “IDx-DR” and will be used to diagnose diabetic retinopathy. It uses an AI enabled algorithm to analyze images of the eye, taken with a retinal camera Topcon NW400. Digital images are uploaded to the cloud server of the device. It can provide screening decision without assistance of a clinician. Out of 900 images of affected eyes, it could correctly identify more than mild diabetic retinopathy with 87% accuracy.

 
  Image  
 

AI Lets Readers “Talk to Books”

Google has introduced ‘Talk to books’ which allows a reader to communicate with a book using a ML trained algorithm that answers to questions from relevant passages of it. It uses ‘natural language understanding’ of words and phrases. After typing in the question, tool searches all the sentences in over 100,000 books to find the ones that respond to that question. It generates a list of plausible responses. Google says that one might find passages that they didn't expect as a response, this serves people discover unexpected authors and titles which they might find relevant.

Information Technology  
  Image  
  Entertainment

AI Created Its Own Flintstones Cartoons

Researchers working at The Allen Institute for Artificial Intelligence, University of Illinois and University of Washington have developed a AI algorithm called “Composition, Retrieval and Fusion Network” (Craft). More than 25,000 Flintstones videos were used as the training set. It uses annotations from videos to determine how the original images correspond to the words used to describe them. With time it makes its own parameters to make individual characters and objects in the cartoon. The developers claim that the algorithm imitates the original cartoon’s style with startling accuracy.

 
  Image  
 

AI to Accelerate Discovery of Metallic Glass

Scientists at the Department of Energy's SLAC National Accelerator Laboratory, the National Institute of Standards and Technology (NIST) and Northwestern University have developed a shortcut for discovering and improving metallic glass with help of AI. Stanford Synchrotron Radiation Lightsource (SSRL) combines machine learning with experiments that quickly make hundreds of sample materials. The team used data of more than 6,000 experiments that searched for metallic glass to train the algorithm. Success rate for finding metallic glass had increased from one out of 300 or 400 samples tested to one out of 1 or 3 samples tested.

Material Science  
  Image  
  Logistics

Start-Up Uses Machine Learning To Predict Train Arrival Time

“RailYatri”, a travel startup, has suggested a technique to predict Estimated Time of Arrival (ETA) using machine learning with precision based on the data. The algorithm they have developed has been trained on historical data of train runs spread over many years. Co-founder Kapil Raizada claims, “its prediction is nearly 110 per cent better than the existing way of estimating train travel time.” The system is open sourced and has already raised funds from investors like Nandan Nilekani, Helion Ventures, Omidyar Partners and Blume Ventures.

 
  Image  
 

An AI Algorithm That Thinks Like A Dog

Kiana Ehsani from University of Washington, Seattle has gathered data set of canine behavior and used it to train an AI algorithm to make dog-like decisions. Data was gathered by putting inertial measurement units on dog’s legs, tail, and body to record their relative angle and its absolute position. Visual scenes and sounds were recorded as well. In total, 24,500 video frames were collected. Out of these, 21,000 were used for training of algorithm and the remaining for validation and testing of the same.

Miscellaneous  
  Image  
 
 
 
     
  Banner  
 
Contact us:

5A, B-Wing, Trade Star Building, J. B. Nagar, Andheri-Kurla Road, Andheri (East), Mumbai - 400 059, Maharashtra, India
W: www.decimalpointanalytics.com  
E: info@decimalpointanalytics.com   T: +91 22300 15200