Ask HN: Is Neuromorphic Computing Promising?
I just read https://arxiv.org/abs/1705.06963. Specifically, the use of organic materials to build robust, low power networks that get past the von Neumann bottleneck and allow us to incorporate new levels of sensing into our environment and lives seems extraordinary. TPUs and chips are obviously interesting for orgs like google, but what about synthetic clusters of neurons integrated into our environment? How significant could that be?
Also, what startups are working on this actively?
2 comments
[ 2.9 ms ] story [ 16.9 ms ] threadYes. Concepts are useless until proven otherwise, but if we want to summarize a whole field of thought and derivative applications, and if we take a reasonable heuristic that a thing can be promising if it shows potential upside without guarantee of success, I would definitively say yes.
For me the phrase is so cool but it lacks sufficient rigor. What's it really mean to take concepts from the brain and apply them, starting with where do you draw the boundary line around member approaches? Neuroscience and computing can inclusively have a lot of overlap: is any logic ‘brain-like’? Do we need to seperate brain and mind and ‘intelligent’ behavior? Your question also begs for a metric of what would be promising. Let's try economically feasible as a filter. There are clear potential benefits to the tech but it's complex and expensive and mostly early r&d, so you see government initiatives, academic projects, niche startups, and FAANG (IBM, Intel). The article citations for this pretty wide field are thorough but here's some personal highlights:
Koniku, biological neural computation start up | koniku.com
blue brain supercomputer sim project | https://www.epfl.ch/research/domains/bluebrain/blue-brain/ab...
Vicarious neuro-centric robotics startup | https://www.vicarious.com/research/
Neuropace anti-epileptic tx device | neuropace.com
Spiking net sim frameworks abound while overshadowed by contemporary ML platforms, I'll feature nengo | nengo.ai but there's also brian, neuron, nest, spike… Nengo is primarily maintained by academic startup ABR | appliedbrainresearch.com/about-us/
Strategic endeavors like Allen Institutes for AI and Brain, Janelia Farms neuroscience research campus, things like Brain Initiative in US and Human Brain Project in Europe, DARPA’s continued interest, plus a number of dedicated world research institutions and departments generate exponential contributions to the state of the art throughout the abstraction stack. This top down money/effort is part of the fermentation tank for future breakthroughs, a parallel to pharmaceutical drug discovery.
Two projects I find interesting if perhaps a lot to live up to Numenta | Numenta.com OpenCog | https://wiki.opencog.org/w/The_Open_Cognition_Project
At the top levels deepmind would define itself as a neuromorphic computing company; youtube interviews with demis hassabis support this. TPU imo is on the spectrum of a neuromorphic computing device. Elon Musk decided he needed to start a neuromorphic interface company, founded an (originally) non profit existential insurance think tank, and aspects of control dynamics and perception for spacex rockets and tesla autonomous cars all arguably fit into the concept of neuromorphic computing.
How far down the rabbit hole do we want to go? Do we need to see direct and immediate impact of r&d or can we value basic research for secondary effects? What's the goal and value prop of any particular research and what side effects might it have? It's conceivable this tech leads to such horrors and wonders as diverse as perfect digital shadows of our consciousness, which we use as back ups or slaves. It could be integral to autonomous killing robots. It could birth the first synthetic consciousness. It could cure your daughter’s disability, or your dementia. It could grow you a chicken breast without the attached chicken. It might give rise to something as menial as an ad algorithm that’s 5% better or profound as collective consciousness or the end of humanity. At this level of abstraction, tech is a double edg...