The human brain, incredibly energy-efficient at 20 watts, processes an exaflop of operations per second. Western Sydney University in Australia is developing DeepSouth, the world's first supercomputer simulating networks at this scale. Expected to operate next year, it aims for 228 trillion synaptic operations, potentially advancing cyborg brain capabilities.
Revolutionizing Understanding of Neural Computation
The objective of DeepSouth is to comprehend how brains efficiently process substantial information with minimal power, potentially leading to the creation of a more potent cyborg brain. This breakthrough could revolutionize our understanding of brain functionality.
André van Schaik, a director at Western Sydney University, notes the hindrance in understanding neural computation due to the inability to simulate brain-like networks at scale.
Van Schaik emphasizes that simulating spiking neural networks on conventional computers using GPUs and multicore CPUs is slow and power-intensive, asserting that their system will address these challenges.
Moreover, Ralph Etienne-Cummings from Johns Hopkins University views DeepSouth as a groundbreaking tool for neuroscience research, particularly for scientists exploring brain function and those developing AI solutions. Etienne-Cummings identifies two key interest groups in this technology: neuroscience researchers and those aiming to prototype AI engineering solutions.
DeepSouth is just one of several projects aspiring to create machines rivaling the human brain. Other researchers are exploring "biological computers" powered by real brain cells. Set to go online in April 2024, DeepSouth aims for high-speed data processing, compact size, and low energy consumption, utilizing a spiking neural network approach.
Its modular and scalable design, using off-the-shelf hardware, allows for future expansion or contraction to adapt to diverse tasks. The overarching goal is to advance AI processing, aligning it more closely with human brain mechanisms and contributing insights relevant to various fields.
How Brains Process Information
The human brain processes information through intricate algorithms in both bottom-up and top-down processing. Sensory input transforms, with bottom-up processing combining simple information to recognize objects, while top-down processing accelerates recognition based on previous decisions.
Attention filters, essential for prioritizing signals, modulate the results of processing stages, allowing selective awareness and focus on relevant information.
Information processing in the brain involves encoding, storing, and retrieving memories. Different memory types, such as sensory, working, and long-term, rely on specific encoding methods for various stimuli. Maintenance of working memory, lasting around 20 seconds, may involve electrical signals looping through specific neurons.
Memory retrieval, not akin to watching recorded videos, relies on environmental retrieval cues, triggering the brain to reconstruct past events based on stored details. However, retrieval cues can lead to memory distortions, emphasizing the complexity and nuances of information processing in the brain.
The organization of knowledge in the brain remains a topic of diverse models, including semantic networks and connectionist models, each with strengths and weaknesses. As research on information processing in the brain continues, its implications extend beyond fascination, with an increasing number turning to psychology for insights into study skills and cognitive enhancement.
RELATED ARTICLE:
Check out more news and information on Supercomputer in Science Times.
