Controlling Computers with Our Thoughts

Described in a recent Microsoft patent application is a mind-reading headset-like device that enables users to control apps and computers using their thoughts. According to the details in the document, the device is capable of reading signals from individual brains and acting on them. This is by detecting neurological intent data from the user. This means that people could soon be able to use their favorite apps completely hands-free. According to Microsoft, if the device works as intended, users can use their thoughts to play games, use web browsers and word processors, interact using VR and use apps to control machinery and other mechanical tools.

The device detects a user’s intent associated with a particular application function and then automatically changes that application’s state to align with the intended operation. In the same way amputee’s think about moving amputated limbs, in some circumstances, neurological user intent data corresponding to a physical gesture will generate that exact gesture. Microsoft does not provide methods to be used by the device to gather people’s neurological data. The technology, however, correlates with functional magnetic resonance imaging (fMRI), magnetoencephalography (MEG) and electroencephalography (EEG). All these methods are non-invasive and are used differently to measure brain activity. Other unspecified methods may also be included.

Research source link:

http://www.independent.co.uk/life-style/gadgets-and-tech/news/mind-reading-headset-computer-control-thoughts-microsoft-patent-a8163976.html

50 Years of Data Collection to Help Computers in Cancer Prognosis

Currently, the tools used by doctors to tell the difference between severe and not-so-severe cancer are not efficient nor precise enough. When a doctor suspects that a specific tissue could be cancerous, he/she takes a sample and holds onto to it, testing to determine if it’s true. In Norway, these samples could stay in labs for months due to lack of enough specialists. Also, prostate cancer is the most common type of cancer among Norwegian men. An 80-year-old Norwegian clinician, Hakon Waehre, specialized in both urology and surgery has been monitoring and collecting data from prostate cancer patients for more than 50 years. Thanks to him, future computers could learn how to recognize cancerous tissue.

To achieve this, lots of data have to be fed into machine learning systems. The Oslo University Hospital launched a research project dubbed DoMore!, which is currently using the latest tools such as Deep Learning and Big Data to automate the process of identifying cancer cells. This needs a lot of data from patients. Computers are fed with vast amounts of data and are trained to recognize patterns. With what Hakon Waehre has, the machines could learn how to differentiate between dangerous and less dangerous cancer cells. While being interviewed, Waehre emphasized the importance of having a properly working abdomen. Researchers are currently working on providing safer and faster systems for cancer prognosis. They aim at achieving this by 2021.

Research source link:

http://sciencenordic.com/old-patient-stories-help-computers-predict-cancer

Security for all Computers Worsens

A group of researchers announced on Wednesday, last week, that there have been serious security vulnerabilities at the core of all computers produced in the last 15 to 20 years. Researchers named these vulnerabilities Meltdown and Spectre. According to them, in order to fix this issue, we would have to throw away our computers and buy new ones that will come out in the next few years. The two vulnerabilities are capable of manipulating the performance of computer processors by changing the order in which instructions are availed to them. An attacker from outside who is capable of taking control of any of these vulnerabilities can steal secrets and vital information while operating from elsewhere.

Additionally, malicious apps and programs on smartphones and computers can easily access other apps and steal data. These are normally obtained from untrustworthy websites and phishing attacks. Among those most vulnerable include cloud services, which share machines amongst several users. Corporates that use cloud infrastructure to store information could be at great risk. The information about these vulnerabilities was kept secret by major IT companies as they sought ways of ramifying and coordinating updates. It, however, leaked before being officially released. Researchers also claim that patching these threats would greatly slow computers, leaving redesigning processors as the only solution. Some solutions may even require consumers to update their firmware, which can easily lead to bricked devices. Researchers claim that they could soon find more vulnerabilities, which are worse than Spectre and Meltdown.

Research source link:

http://www.cnn.com/2018/01/04/opinions/security-of-nearly-every-computer-has-just-gotten-a-lot-worse-opinion-schneier/index.html

Teaching Machines to Think Like Humans

Picture a machine that can predict words before they are said. One which can think like a human being and predict future outcomes based on present circumstances. A new neural network dubbed reservoir computing system has been created by a team of researchers led by University of Michigan’s very own professor of computer science and electrical engineering, Wei Lu. The researchers published their work in the Nature Communications journal. The new type of network greatly improves the teaching time required to train machines. In the past, scientists have improved typical neural networks to shorten training periods using larger optical components. What makes this research different is that Lu and his colleagues used memristors.
These are special types of resistive devices that can store data and perform logical operations. Regular computers have these two aspects separated and carried out independently. The memristor used can store data based on events that occurred in the recent past. During training, neural networks are normally subjected to questions and answers over a period of time. This is constantly monitored to minimize errors. Once done, the network is tested on a new question without knowing the answer. For example, a neural network can correctly identify someone from a photo since it has learned how to do this through training. Training may take months and is often very expensive. Lu hopes to train the new neural networks to produce clean data from noise signals from afar.
Research source link:
https://www.sciencedaily.com/releases/2017/12/171222090313.htm