Connect with us

Deep Learning

New Tool Can Show Researchers What GANs Leave Out Of An Image

mm

Published

 on

New Tool Can Show Researchers What GANs Leave Out Of An Image

Recently, a team of researchers from the MIT-IBM Watson AI Lab created a method of displaying what a Generative Adversarial Network leaves out of an image when asked to generate images. The study was dubbed Seeing What a GAN Cannot Generate, and it was recently presented at the International Conference on Computer Vision.

Generative Adversarial Networks have become more robust, sophisticated, and widely used in the past few years. They’ve become quite good at rendering images full of detail, as long as that image is confined to a relatively small area. However, when GANs are used to generate images of larger scenes and environments, they tend not to perform as well. In scenarios where GANs are asked to render scenes full of many objects and items, like a busy street, GANs often leave many important aspects of the image out.

According to MIT News, the research was developed in part by David Bau, a graduate student at the Department of Electrical Engineering and Computer Science at MIT. Bau explained that researchers usually concentrate on refining what machine learning systems pay attention to and discerning how certain inputs can be mapped to certain outputs. However, Bau also explained that understanding what data is ignored by machine learning models if often just as important and that the research team hopes their tools will inspire researchers to pay attention to the ignored data.

Bau’s interest in GANs was spurred by the fact that they could be used to investigate the black-box nature of neural nets and to gain an intuition of how the networks might be reasoning. Bau previously worked on a tool that could identify specific clusters of artificial neurons, labeling them as being responsible for the representation of real-world objects such as books, clouds, and trees. Bau also had experience with a tool dubbed GANPaint, which enables artists to remove and add specific features from photos by using GANs. According to Bau, the GANPaint application revealed a potential problem with the GANs, a problem that became apparent when Bau analyzed the images. As Bau told MIT News:

“My advisor has always encouraged us to look beyond the numbers and scrutinize the actual images. When we looked, the phenomenon jumped right out: People were getting dropped out selectively.”

While machine learning systems are designed to extract patterns from images, they can also end up ignoring relevant patterns. Bau and other researchers experimented with training GANs on various indoor and outdoor scenes, but in all of the different types of scenes the GANs left out important details in the scenes like cars, road signs, people, bicycles, etc. This was true even when the objects left out were important to the scene in question.

The research team hypothesized that when the GAN is trained on images, the GAN may find it easier to capture the patterns of the image that are easier to represent, such as large stationary objects like landscapes and buildings. It learns these patterns over other, more difficult to interpret patterns, such as cars and people. It has been common knowledge that GANs often omit important, meaningful details when generating images, but the study from the MIT team may be the first time that GANs have been demonstrated omitting entire object classes within an image.

The research team notes that it is possible for GANs to achieve their numerical goals even when leaving out objects that humans care about when looking at images. If images generated by GANS are going to be used to train complex systems like autonomous vehicles, the image data should be closely scrutinized because there’s a real concern that critical objects like signs, people, and other cars could be left out of the images. Bau explained that their research shows why the performance of a model shouldn’t be based only on accuracy:

“We need to understand what the networks are and aren’t doing to make sure they are making the choices we want them to make.”

Spread the love

Deep Learning Specialization on Coursera

Blogger and programmer with specialties in machine learning and deep learning topics. Daniel hopes to help others use the power of AI for social good.

Deep Learning

A New AI System Could Create More Hope For People With Epilepsy

mm

Published

on

A New AI System Could Create More Hope For People With Epilepsy

As Endgadget reports, two AI researchers may have created a system that creates new hope for people suffering from epilepsy – a system “that can predict epileptic seizures with 99.6-percent accuracy,” and do it up to an hour before seizures occur.

This would not be the first new advancement, since previously researchers at Technical University (TU) in Eindhoven, Netherlands developed a smart arm bracelet that can predict epileptic seizures during nighttime. But the accuracy and short time-frame the new AI system can work on as IEEE Spectrum notes, gives more hope to around 50 million people around the world who suffer from epilepsy (based on the data from World Health Organization). Out of this number of patients, 70 percent of them can control their seizures with medication if taken on time.

The new AI system was created by Hisham Daoud and Magdy Bayoumi of the University of Louisiana at Lafayette, and is lauded as “a major leap forward from existing prediction methods.” As Hisham Daoud, one of the two researchers that developed the system explains, “Due to unexpected seizure times, epilepsy has a strong psychological and social effect on patients.”

As is explained, “each person exhibits unique brain patterns, which makes it hard to accurately predict seizures.” So far, the previously existing models predicted seizures “ in a two-stage process, where the brain patterns must be extracted manually and then a classification system is applied,” which, as Daoud explains, added to the time needed to make a seizure prediction.

In their approach explained in study published on 24 July in IEEE Transactions on Biomedical Circuits and Systems, “the features extraction and classification processes are combined into a single automated system, which enables earlier and more accurate seizure prediction.”

To further boost the accuracy of their system Daoud and Bayoumi “incorporated another classification approach whereby a deep learning algorithm extracts and analyzes the spatial-temporal features of the patient’s brain activity from different electrode locations, boosting the accuracy of their model.” Since “EEG readings can involve multiple ‘channels’ of electrical activity,” to speed up the prediction process, even more, the two researchers “applied an additional algorithm to identify the most appropriate predictive channels of electrical activity.”

The complete system was then tested on 22 patients at the Boston Children’s Hospital. While the sample size was small, the system proved to be very accurate (99.6%), and had “a low tendency for false positives, at 0.004 false alarms per hour.”

As Daoud explained the next step would be the development of a customized computer chip to process the algorithms.  “We are currently working on the design of efficient hardware [device] that deploys this algorithm, considering many issues like system size, power consumption, and latency to be suitable for practical application in a comfortable way to the patient.”

Spread the love

Deep Learning Specialization on Coursera
Continue Reading

Deep Learning

DeepMind’s AI Reaches Highest Rank of StarCraft II

Published

on

DeepMind’s AI Reaches Highest Rank of StarCraft II

DeepMind’s AlphaStar, an artificial intelligence (AI) system, has reached the highest level in StarCraft II, an extremely popular and complex computer game. The AI outperformed 99.8% of all registered human players.

It took the AI system 44 days of training to be able to reach the level. It used recordings of some of the best human players, and it learned from them until eventually going up against itself. 

“AlphaStar has become the first AI system to reach the top tier of human performance in any professionally played e-sport on the full unrestricted game under professionally approved conditions,” said David Silver, a researcher at DeepMind.

“Ever since computers cracked Go, chess and poker, the game of StarCraft has emerged, essentially by consensus from the community, as the next grand challenge for AI,” Silver said. “It’s considered to be the game which is most at the limit of human capabilities.”

The work was published in the scientific journal Nature

What is StarCraft?

Put simply, the point of StarCraft is to build civilizations and fight against aliens. 

It is a real-time strategy game where players control hundreds of units and have to make important economic decisions. Within the game, there are tens of thousands of time-steps and thousands of possible actions. These are selected in real-time throughout ten minutes of gameplay. 

AlphaStar “Agents”

DeepMind developed AlphaStar “Agents,” and they created one for each of the different races in the game. The different races each have a unique set of strengths and weaknesses. In the “AlphaStar league,” the AI competed against itself and “exploiter” agents which targeted the weaknesses of AlphaStar. 

One of the most impressive points of the AI was that it was not developed to perform actions at superhuman speed. Instead, it learned different winning strategies. 

Just like the StarCraft game, real-world applications require artificial agents to interact, compete, and coordinate within a complex environment containing other agents. This is why StarCraft has become such an important aspect of artificial intelligence research. 

Military Interest

Perhaps one of the more unexpected aspects of this work is that it’ll be of interest to the military. 

“Military analysts will certainly be eyeing the successful AlphaStar real-time strategies as a clear example of the advantages of AI for battlefield planning. But this is an extremely dangerous idea with the potential for humanitarian disaster. AlphaStar learns strategy from big data in one particular environment. The data from conflicts such as Syria and Yemen would be too sparse to be of use,” said Noel Sharkey, a professor of AI and robotics at the University of Sheffield.

“And as DeepMind explained at a recent United Nations event, such methods would be highly dangerous for weapons control as the moves are unpredictable and can be creative in unexpected ways. This is against the laws that govern armed conflict.”

Coming a Long Way in Short Time

Back in January, professional StarCraft II player Grzegorz Komincz, defeated AlphaStar in the game. It was a huge set back for Google, who had invested millions of dollars into the technology. Since then, DeepMind’s AI has come a long way in a short amount of time, and these new developments have huge implications.

Spread the love

Deep Learning Specialization on Coursera
Continue Reading

Deep Learning

Researchers Train An AI To Predict The Smell Of Chemicals

mm

Published

on

Researchers Train An AI To Predict The Smell Of Chemicals

A recent paper published by researchers at Google Brain demonstrates how researchers managed to train an AI to predict the smell of objects, based on the structure of the chemicals passed into the network. As reported by Wired, the researchers are hopeful that their work could help unravel some of the mysteries surrounding the human sense of smell, which is poorly understood in comparison to our other senses.

The differences between smells are complex and a single atom being changed in a molecule can change a smell from pleasant to unpleasant. It’s difficult for researchers to understand the patterns that cause chemical structures to be interpreted by our olfactory senses as pleasant or aversive. In contrast, the patterns of the electromagnetic spectrum that appear as color to our eyes are much more easily quantifiable, with scientists being able to make precise measurements that will tell them what certain wavelengths of light will look like.

Machine learning algorithms excel at finding patterns within data, and for this reason, AI researchers have attempted to use machine learning to gain better insight into how smells are interpreted by the human brain. Attempts to utilize machine learning algorithms to quantify smell include the DREAM Olfaction Prediction Challenge carried out in 2015. Several studies took the data from the challenge and tried to generate natural language descriptions of mono-molecular odorants.

The recent study, published in Arxiv, catalogs the Google Brain researcher’s attempts to quantify smell using neural networks. The researchers utilized a Graph Neural Network or GNN. Graph Neural Networks are capable of interpreting graph data, which are data structures comprised of nodes and edges. Graphs are commonly used to represent networks or relationships between individual data points. In the context of a social network, a graph would have each person in the network represented by a node or vertex. Such graphs are used by social media companies to predict people on the peripherals of your current network and suggest new friends.

For the purposes of interpreting smells, the researchers trained the network on thousands of molecules, each matched with a natural language descriptor. The GNN was able to interpret the data and pick up on patterns in the structure of the molecules. The descriptors used by the researchers were phrases like “sweet”, “smoky”, or “woody”. Approximately two-thirds of the over 5,000 molecules that were compiled by the researchers were used to train the model, while the remaining third was used to test the model.

The model the researchers trained worked so well that once the first iteration was completed, performance already matched the peak performance achieved by other groups of researchers who tried to assign natural language labels to chemical structures.

Alex Wiltschko, one of the researchers who worked on the project, acknowledges that there are a couple of limitations to their current approach. For one, the AI may distinguish differences between chemical structures that humans would describe as being the same, calling two different chemicals “earthy” or “woody” in nature, even though the AI classifies them differently. Another issue with the classifier is that it doesn’t distinguish between chiral pairs, which are molecules that are mirror images of each other. The different orientations mean they have different smells, but the model currently doesn’t see them as being different.

The research team aims to address these limitations in their future work. The research still has a long way to go, but it is a step towards understanding what features of a molecule correspond with our perception of certain smells. The Google Brain team isn’t the only research team to be working on applications of AI aimed at recognizing scents. Other AI experiments involving scent include IBM’s experiments with AI-generated perfumes and an experiment by Russian scientists to detect potentially toxic mixtures of gas.

Spread the love

Deep Learning Specialization on Coursera
Continue Reading