🧠 Today we introduce Un-0 from @unconvAI : the first large-scale generative model build on physics as a compute primitive. This represents a “hello world” moment for physics-based models. We use the inherent time-varying behavior of physical systems to do compute for us. The result is a new way to build a computer that can be VASTLY more power efficient. 🧵 https://unconv.ai/blog/introducing-un-0-generating-images-with-coupled-oscillators/
Many users expressed excitement about UnconvAI's Un-0 physics-based generative model using oscillators, praising the novel approach and fast shipping, while one found the announcement lacking substance.
No Digg Deeper questions have been answered for this story yet.
How does this relate to energy efficiency? Most energy in existing von Neumann machines goes into moving information between memory and compute elements. Dynamical systems combine compute and memory into a single entity. What’s more, dynamical systems can tolerate noise. This opens up new opportunities to save energy in communication even further.
Why is this significant? It shows that computing is not some unique invention from humans; it’s present all through nature and physics. All physics of all physical entities has time; however, current computing systems don’t. We’re exploiting that time dimension.
Why is this significant? It shows that computing is not some unique invention from humans; it’s present all through nature and physics. All physics of all physical entities has time; however, current computing systems don’t. We’re exploiting that time dimension.
Un-0 represents a big first step in changing the paradigm of compute to dynamical systems. We are connecting intelligence to dynamics with this model release. Dynamics are a natural framing for compute for AI; neural networks themselves are really dynamical systems so the mapping becomes more straightforward. The brain does not have an abstraction of linear algebra; so in effect, we’re cutting out the middleman.
How does this relate to energy efficiency? Most energy in existing von Neumann machines goes into moving information between memory and compute elements. Dynamical systems combine compute and memory into a single entity. What’s more, dynamical systems can tolerate noise. This opens up new opportunities to save energy in communication even further.
Churchill quote that works pretty well here (h/t CFO Ali) ... "Now this is not the end. It is not even the beginning of the end. But it is, perhaps, the end of the beginning." We are embarking on a fantastic journey 🚀
@NaveenGRao @unconvAI How do you do credit assignment during training?
@aesfahani
@mcarbin @NaveenGRao @unconvAI Thanks. If you move to physical systems, will you still need to train in simulation?
@SilverJacket @NaveenGRao @unconvAI We train Un-0 with plain ol' backprop
@NaveenGRao @unconvAI Maybe the question isn’t whether nature computes. Maybe it’s how we harness and control those computations in a reliable way.
@NaveenGRao @unconvAI In a physical system, how would you tune the parameters, especially the coupling strengths?
@NaveenGRao @unconvAI Thank you for citing Arthur Winfree. 😉
fascinating! We also discovered that using coupled oscillators works surprisingly well for raw speech recognition. Not entirely sure why but seems like the network can *learn* to exploit the physics of the oscillations (e.g, synchronization, phase coupling, etc.): https://www.sciencedirect.com/science/article/pii/S2666389926000723
@NaveenGRao @unconvAI new physics wave
@NaveenGRao @unconvAI using physics as compute is genuinely sick. this is the kind of weird i want to see more of
@NaveenGRao @unconvAI shipping fast, love to see it!
oh, great questions! The are various different strategies for this that are in the community; and that we are thinking about as well.
Training on GPUs and deploying on the physical fabric is possible. Also, training on GPU and finetuning on physical. Or even training from scratch on physical. It depends on the capabilities physical fabric itself.
@NaveenGRao @unconvAI 🇺🇬 Ugandan female artist, I'm so excited 😊 coz I just released my song http://ditto.fm/post-me-oranda OUT NOW 🤭
"dynamical systems combine compute and memory into a single entity."
The main challenge for deep learning acceleration lies in the frequent data transfer between compute units and memory in the conventional von Neumann architecture, with data movement incurring up to 1000 times more energy consumption due to frequent movement of data between memory and processors.
@NaveenGRao cool, but you said a lot without saying anything!
@NaveenGRao @unconvAI New computing paradigms could reshape AI efficiency. @Yellow provides the trust and settlement infrastructure needed as autonomous AI systems become more capable.
