AI Convolutional Neural Network: Understanding Volcanic Activity

Volcanic explosions are displayed in different forms. Volcanologists have recently discovered an advanced computer algorithm to help understand volcanic eruptions deeper. The artificial intelligence called Convolutional Neural Network or CNN is designed to analyze volcanic ash particles accurately.

Unlike the existing programs, CNN do not only perform simple tests, but it also rapidly learns on its own. The traditional method of ash particle classification using the naked eye is time consuming and often subjective. The process is also slowed down by limited number of expert volcanologists. CNN offers researchers advanced tools that would help accurately identify what type of eruption occurred.

Convolutional Neural Network is a useful tool that makes complex data identification easier. The program can also be modified to help broaden its functionality, such as the addition of Electron Microscopy to aid increased visual magnification, and make particle identification easier by adding color.

CNN opened the door for possibilities and researchers are hoping to discover the unknown in the world of particle microscopy.

Reference: https://www.sciencedaily.com/releases/2018/06/180628171415.htm

Predicting Future with a Computer Program

People have said that future is unpredictable and can change depending on present actions. However, this is about to change. Now, Professor Dr. Jürgen Gall wants the computers to predict the timing and duration of activities ahead of time. For example, a kitchen robot can pass ingredients, preheat ovens, or warn the chef that he or she is about to miss a step of preparation for a dish. Also, an automatic vacuum cleaner can clean other places in the house until the cooking ends in the kitchen.

As of now, the computer programs are not developed enough to predict actions like humans can. The University of Bonn was able to successfully allow the self-learning program to estimate the duration and timing of future activities for several minutes.

The computer program “watched” four hours worth of different salad preparation videos that contained multiple different actions. This allowed the algorithm to learn the typical time that it takes to prepare a bowl of salad. However, this depends on the chef. Thus, further learning for the program is needed. Despite the accuracy of the program is lower than 50%, further research and “learning” for the program can further increase the accuracy of the computer program.

Source: https://www.sciencedaily.com/releases/2018/06/180613102016.htm

 

RSA Encryption

RSA encryption is one of the most commonly used public encryption methods today. RSA encryption contains elegance and complication of number theory. Initially, the encryption method had been classified, however private individuals have rediscovered the encryption method, thus becoming public. The function of the RSA encryption can first seem confusing, but when comprehended the beauty can be seen.

For example, Sammy wants to send a secure message to Mark. However, Dan is always intercepting the message when sent. Thus, Mark and Sammy decide to encrypt the message so Dan can’t see the private message. First, Sammy converts his message, “hi”, into numbers using a padding scheme, and the result was 89. Let’s call this number m. Now it is Mark’s turn to start the encryption. Mark chooses two prime numbers of similar size. He chooses P1, the first prime number, as 23 and P2, the second prime number, as 19. Then, he calculates the product of P1 and P2 that results as 437. Let’s call this number n.

Now, Mark needs to use a Phi function. Phi function tests the breakability of a number. ϕ(n) — where n is a prime number — is always n-1. Also, ϕ(P1*P2) is always equal to ϕ(P1)*ϕ(P2). n for Mark is 437, and we know both of the prime numbers that Mark used: 23 and 19. Thus, ϕ(437)=ϕ(23)*ϕ(19) which is 22*18=396. Now we need to choose e, a number that is odd and not a factor of ϕ(n). In this case, e can be 7. Now, we are going to set a key called d. The equation for this key is (k*ϕ(n)+1)/e. The integer k must make the quotient integer and can be any kind of positive integer. In Mark’s case, k is 5. Now, when everything is plugged in, the result of d is 283.

Now, Mark hides everything except n and e. This way, when Dan intercepts the message sent by Mark, he only gets the information on n and e. Sammy received the encrypted numbers from Mark, and now it is his turn to encrypt the message he wants to send to Mark. The function that Sammy uses is (m^e) mod (n). Let’s call the resulting value as c. When the numbers are plugged in, the result for c will be 67. Now, Sammy sends c to Mark for decryption. The only information that Dan has is n, e, and c.

For decryption, Mark uses the function c^d mod n to see what m is. This is because c^d = m mod n, and m^(k*ϕ(n)+1) = m mod n which derives from m^(k*ϕ(n)) = 1 mod n. Thus, Mark plugs in the numbers and get (67^283) mod (437). This will decrypt the message and allow Mark to see the number 89 without Dan knowing what the secret message is.

Unless Dan knows the two prime numbers chosen by Mark, it will take Dan lots of time to figure out the factors by brute force method. The example used uses a small number, however, if the two prime numbers chosen by Mark is extremely big, then it will take hundreds or thousands of years to figure out the factor of n. This is what keeps the RSA encryption secure. Today’s computer cannot crack the RSA encryption in a reasonable amount of time because of the beauty of prime number. There are an infinite amount of prime numbers, and the only method that is the fastest is the brute force method only. Modern encryption uses the beauty of number theory, and it may seem to be hard to comprehend everything. However, all online communications are safe because of these type of encryption.

Source: https://www.khanacademy.org/computing/computer-science/cryptography/modern-crypt/v/rsa-encryption-part-4

*(All the videos from the Modern Cryptography section in Khan Academy are used)*

AI Peeks Five Minutes Ahead of Time

An artificial intelligence that can predict future actions with precision was created by researchers from the University of Bonn and is currently in its infancy. The software learns the pattern of actions through hours of watching specified videos with detailed information.

Once the algorithm watched the training data, it learns what the next action will be. Upon testing, the software was provided with new footages in which it reaped forty percent accuracy on short-term prediction. When the system was confronted with more than three minutes of a forecast, its precision dropped to fifteen percent, considering that the ground of success is based on proper activity forecast and its timing.

Noticeably, the system’s accuracy plummets when it is asked to analyze the next action on its own instead of being fed with data. Dr. Jȕrgen Gall and his team are continuing deeper into the study to improve the algorithm’s analyzing skills and expecting a future where AI’s can predict actions hours ahead with precision.

Reference: https://www.sciencedaily.com/releases/2018/06/180613102016.htm

Artificial Intelligence Can Identify, Count, and Describe Wild Animals

The AI today can identify the animal by motion-sensor cameras. The AI can identify animal up to 99.3% of the image and perform with a 99.6% accuracy. The AI can aid biologists, zoologists, ecologists, and more by storing the big data that is crowdsourced from humans. This ability can improve the study and conservation of wildlife and ecosystems.

The Deep Neural Network, what the AI uses to describe the animal, is a form of computational intelligence inspired by the function of animal brains and the way animals see the world. This requires an immense amount of data and training of the AI to be accurate.

The pictures were taken by motion motion-sensor camera then converted into text and numbers for the AI to process and store the data. The best way to be accurate with data was crowdsourcing the data with human help and label every image. Since there are thousands of species of animals, the algorithm in the AI program is expected to be able to process data faster than human brains like many other data processed today.

As of now, the AI is able to label 48 different species, count how many there are, and describe the action done by the animals. The presence of baby animals, animals eating, and animals sleeping can all be described by the AI. The AI is expected to decrease eight years of human labeling efforts for each 3 million images. The AI is yet to have its full potential to be seen, but with continuous training, the day will come soon.

Source: https://www.sciencedaily.com/releases/2018/06/180605124148.htm

AI Privacy Filter

Researchers from U of T Engineering have recently created an algorithm as a privacy filter for photos. It alters pictures by disrupting facial recognition systems built within internet platforms. The whole system is run by two AI algorithms: one which performs face detection and the other created to oppose the first process.

Internet privacy concerns are alarming, especially that facial recognition tools are becoming more advanced. Professor Parham Aarabi and Avishek Pose developed and trained AIs to compete constantly, countering each other’s performances. This resulted in a photo filter that specifically targets to manipulate certain pixels within a photo without totally distorting the image. Additionally, this tool also disables any image search engines from extracting results.

The fascinating algorithm can learn on its own through systematic repetition of data and continuously improve its performance via competition. The researchers are looking forward into making the tool available to the public through app or website.

Reference: https://www.sciencedaily.com/releases/2018/05/180531114620.htm

AI Analyzing Gut Microbial Patterns

The International Center for Diarrheal Disease Research at Dhaka, Bangladesh, Massachusetts General Hospital and Duke University’s researchers has recently developed an algorithm that detects patterns of bacteria within human body’s gut that are invisible to doctor’s vision.

Exploiting AI advancements, researchers used machine learning tools to develop a system focusing on detection of gut microbe patterns. This is a vital key towards progress of vaccines and other preventive methods against cholera and contagious diseases. Predictive “microbiota” (a community of trillions of gut bacteria) accurately shows cholera risk percentage and allowed the medical field to identify unknown cholera rick factors.

In traditional analysis, gut bacteria are studied individually, taking a lot of time and effort. The machine was trained through sequenced analysis of over 4000 various germ taxa per swab sample, searching patterns that show differences between low and high risk of cholera infection, resulting on identification of 100 microbes that contributes to high risks of cholera.

This AI’s ability to scan hundreds of species in a single run helps prevent the infection beforehand, giving the medical world a clearer view of gut bacteria’s role in diarrheal infections.

Reference: https://www.sciencedaily.com/releases/2018/05/180507153127.htm