A New Approach to Computation: Hyperdimensional Computing

A New Approach to Computation: Hyperdimensional Computing

In the world of artificial intelligence (AI), the success of large language models like ChatGPT has been undeniable. However, some experts argue that the underlying artificial neural networks (ANNs) might be on the wrong track. These systems are so complex that nobody truly understands how they work or why they are so effective. This complexity also hinders their ability to reason by analogy, a fundamental aspect of human cognition.

Neuroscientists, such as Bruno Olshausen from the University of California, Berkeley, propose a different perspective. They suggest that the information in the brain is represented by the activity of numerous neurons, rather than individual neurons encoding specific concepts. For example, the perception of a purple Volkswagen would not be encoded by a single neuron, but rather by the collective activity of thousands of neurons. These same neurons, when firing differently, could represent an entirely different concept, such as a pink Cadillac.

This idea forms the basis of hyperdimensional computing, a radically different approach to computation. In hyperdimensional computing, each piece of information is represented as a hyperdimensional vector, where the concept of a car or its make, model, or color are all part of a single entity. The advantage of this representation lies in the ease of creating nearly orthogonal vectors, allowing for efficient computation.

Hyperdimensional computing also offers advantages in terms of hardware. It is well-suited for low-power and sturdy hardware, making it compatible with in-memory computing systems. Unlike traditional von Neumann computers, which shuttle data between memory and the central processing unit, in-memory computing systems perform computations on the same hardware that stores data. Additionally, hyperdimensional computing can overcome the limitations imposed by random noise in analog devices that operate at low voltages.

Despite its potential, hyperdimensional computing is still in its infancy. It needs to be tested against real-world problems and scaled up to match the size of modern neural networks. Efficient hardware is also a requirement for hyperdimensional computing to reach its full potential, especially when dealing with large-scale problems such as searching over billions of items.

However, researchers remain optimistic about the future of hyperdimensional computing. According to Pentti Kanerva, a researcher at the Redwood Center for Theoretical Neuroscience, high-dimensional spaces hold many other secrets that are yet to be discovered. This represents just the beginning of a new era in computing with vectors.

Actionable Advice:

• 1. Explore hyperdimensional computing: Familiarize yourself with the concepts and principles of hyperdimensional computing. Stay updated on the latest research and advancements in this field.

• 2. Invest in efficient hardware: As hyperdimensional computing evolves, it will require efficient hardware to reach its full potential. Stay informed about advancements in hardware technologies that can support hyperdimensional computing.

• 3. Test hyperdimensional computing on real-world problems: As hyperdimensional computing is still in its early stages, it is essential to test its efficacy on real-world problems. Consider collaborating with researchers and experts in the field to explore how hyperdimensional computing can address specific challenges in your industry.

In conclusion, hyperdimensional computing offers a new approach to computation that challenges the traditional paradigm of artificial neural networks. By representing information as hyperdimensional vectors, it allows for more efficient computation and opens up possibilities for low-power, sturdy hardware. While still in its infancy, hyperdimensional computing holds great potential for solving complex problems and pushing the boundaries of AI. With time, further research, and advancements in hardware, hyperdimensional computing could revolutionize the field of artificial intelligence.