Quanta The magazine believes there are better alternatives to artificial neural networks (ANNs) that power AI systems. (Alternate URL)
As an example, Cornelia Vermüller, a computer scientist at the University of Maryland, said ANN is “extremely power-hungry.” “And another problem is that [their] Systems like this are so complex that no one really understands what they are doing or why they work so well. This makes it nearly impossible for the system to reason by the analogies humans make. — Use symbols for objects, ideas and the relationships between them.
Neuroscientist Bruno Olshausen of the University of California, Berkeley and others argue that information in the brain is represented by the activity of large numbers of neurons… This is known as hyperdimensional computing, a fundamental is a starting point for different computational approaches. The point is that each piece of information, such as car concept, make, model, color, or all, is represented as a single entity: a hyperdimensional vector. A vector is simply an ordered array of numbers. For example, a 3D vector he consists of three numbers, the x, y, z coordinates of a point in 3D space. A hyperdimensional vector (hypervector) can be, for example, an array of 10,000 numbers representing points in a 10,000-dimensional space. These mathematical objects and the algebra that manipulates them are flexible and powerful enough to push modern computing beyond some of its current limitations and to facilitate new approaches to artificial intelligence. …
In hyperdimensional computing, even if a hypervector undergoes a fair number of random bit-flips, it is still close to the original vector and thus more tolerant of errors. This means that inferences using these vectors are not meaningfully affected in the face of errors. Villanova University computer scientist Xun Jiao’s team found that these systems were at least 10 times more tolerant of hardware failures than his traditional ANNs, and that the ANNs themselves were better suited for traditional computing than his architecture. It has shown to be an order of magnitude more resilient.
All these advantages over conventional computing suggest that hyperdimensional computing is suitable for a new generation of extremely robust and low-power hardware. It is also compatible with “in-memory computing systems,” which run computing on the same hardware that stores data (the existing von Neumann computer). Some of these new devices are analog and operate at very low voltages, making them energy efficient but prone to random noise.
Thanks to Slashdot reader ZipNada for sharing the article.
