Nav: Home

Analytic technique could allow neural networks to run on cellphones

July 18, 2017

In recent years, the best-performing artificial-intelligence systems -- in areas such as autonomous driving, speech recognition, computer vision, and automatic translation -- have come courtesy of software systems known as neural networks.

But neural networks take up a lot of memory and consume a lot of power, so they usually run on servers in the cloud, which receive data from desktop or mobile devices and then send back their analyses.

Last year, MIT associate professor of electrical engineering and computer science Vivienne Sze and colleagues unveiled a new, energy-efficient computer chip optimized for neural networks, which could enable powerful artificial-intelligence systems to run locally on mobile devices.

Now, Sze and her colleagues have approached the same problem from the opposite direction, with a battery of techniques for designing more energy-efficient neural networks. First, they developed an analytic method that can determine how much power a neural network will consume when run on a particular type of hardware. Then they used the method to evaluate new techniques for paring down neural networks so that they'll run more efficiently on handheld devices.

The researchers describe the work in a paper they're presenting next week at the Computer Vision and Pattern Recognition Conference. In the paper, they report that the methods offered as much as a 73 percent reduction in power consumption over the standard implementation of neural networks, and as much as a 43 percent reduction over the best previous method for paring the networks down.

Energy evaluator

Loosely based on the anatomy of the brain, neural networks consist of thousands or even millions of simple but densely interconnected information-processing nodes, usually organized into layers. Different types of networks vary according to their number of layers, the number of connections between the nodes, and the number of nodes in each layer.

The connections between nodes have "weights" associated with them, which determine how much a given node's output will contribute to the next node's computation. During training, in which the network is presented with examples of the computation it's learning to perform, those weights are continually readjusted, until the output of the network's last layer consistently corresponds with the result of the computation.

"The first thing we did was develop an energy-modeling tool that accounts for data movement, transactions, and data flow," Sze says. "If you give it a network architecture and the value of its weights, it will tell you how much energy this neural network will take. One of the questions that people had is 'Is it more energy efficient to have a shallow network and more weights or a deeper network with fewer weights?' This tool gives us better intuition as to where the energy is going, so that an algorithm designer could have a better understanding and use this as feedback. The second thing we did is that, now that we know where the energy is actually going, we started to use this model to drive our design of energy-efficient neural networks."

In the past, Sze explains, researchers attempting to reduce neural networks' power consumption used a technique called "pruning." Low-weight connections between nodes contribute very little to a neural network's final output, so many of them can be safely eliminated, or pruned.

Principled pruning

With the aid of their energy model, Sze and her colleagues -- first author Tien-Ju Yang and Yu-Hsin Chen, both graduate students in electrical engineering and computer science -- varied this approach. Although cutting even a large number of low-weight connections can have little effect on a neural net's output, cutting all of them probably would, so pruning techniques must have some mechanism for deciding when to stop.

The MIT researchers thus begin pruning those layers of the network that consume the most energy. That way, the cuts translate to the greatest possible energy savings. They call this method "energy-aware pruning."

Weights in a neural network can be either positive or negative, so the researchers' method also looks for cases in which connections with weights of opposite sign tend to cancel each other out. The inputs to a given node are the outputs of nodes in the layer below, multiplied by the weights of their connections. So the researchers' method looks not only at the weights but also at the way the associated nodes handle training data. Only if groups of connections with positive and negative weights consistently offset each other can they be safely cut. This leads to more efficient networks with fewer connections than earlier pruning methods did.
-end-
Additional background

PAPER: Designing energy-efficient convolutional neural networks using energy-aware pruning

https://arxiv.org/pdf/1611.05128.pdf

ARCHIVE: Peering into neural networks

http://news.mit.edu/2017/inner-workings-neural-networks-visual-data-0630

ARCHIVE: New system allows optical "deep learning"

http://news.mit.edu/2017/new-system-allows-optical-deep-learning-0612

ARCHIVE: Explained: Neural networks

http://news.mit.edu/2017/explained-neural-networks-deep-learning-0414

ARCHIVE: Energy-friendly chip can perform powerful artificial-intelligence tasks

http://news.mit.edu/2016/neural-chip-artificial-intelligence-mobile-devices-0203

Massachusetts Institute of Technology

Related Electrical Engineering Articles:

The effectiveness of electrical stimulation in producing spinal fusion
Researchers from The Johns Hopkins University School of Medicine performed a systematic review and meta-analysis of published data on the effect of electrical stimulation therapies on spinal fusion.
Fat pumps generate electrical power
A previously unknown electrical current develops in the body's cells when the vital fat pump function of the flippases transfers ('flips') lipids from the outer to the inner layer of the body's cell membranes.
UCI electrical engineering team develops 'beyond 5G' wireless transceiver
An end-to-end transmitter-receiver created by engineers in UCI's Nanoscale Communication Integrated Circuits Labs, is a 4.4-millimeter-square silicon chip that is capable of processing digital signals with significantly greater speed and energy efficiency because of its unique digital-analog architecture.
How electrical stimulation reorganizes the brain
Recordings of neural activity during therapeutic stimulation can be used to predict subsequent changes in brain connectivity, according to a study of epilepsy patients published in JNeurosci.
Energy monitor can find electrical failures before they happen
A new system devised by researchers at MIT can monitor the behavior of all electric devices within a building, ship, or factory, determining which ones are in use at any given time and whether any are showing signs of an imminent failure.
More Electrical Engineering News and Electrical Engineering Current Events

Best Science Podcasts 2019

We have hand picked the best science podcasts for 2019. Sit back and enjoy new science podcasts updated daily from your favorite science news services and scientists.
Now Playing: TED Radio Hour

Erasing The Stigma
Many of us either cope with mental illness or know someone who does. But we still have a hard time talking about it. This hour, TED speakers explore ways to push past — and even erase — the stigma. Guests include musician and comedian Jordan Raskopoulos, neuroscientist and psychiatrist Thomas Insel, psychiatrist Dixon Chibanda, anxiety and depression researcher Olivia Remes, and entrepreneur Sangu Delle.
Now Playing: Science for the People

#537 Science Journalism, Hold the Hype
Everyone's seen a piece of science getting over-exaggerated in the media. Most people would be quick to blame journalists and big media for getting in wrong. In many cases, you'd be right. But there's other sources of hype in science journalism. and one of them can be found in the humble, and little-known press release. We're talking with Chris Chambers about doing science about science journalism, and where the hype creeps in. Related links: The association between exaggeration in health related science news and academic press releases: retrospective observational study Claims of causality in health news: a randomised trial This...