Manifold
A lower-dimensional surface embedded in a high-dimensional space; here, the idea that complex natural data actually lies on a simpler learnable structure.
As defined in Demis Hassabis #2
Other definitions
A junction or connection block that distributes fluid or electrical connections from one source to multiple destinations
Why Do Mainframes Still Exist? What's Inside One? 40TB, 200+ Cores, AI, and more!
Where it comes up · 6
- 251 | Rosemary Braun on Uncovering Patterns in Biological Complexity3 claims
- Demis Hassabis #22 claims
- 280 | François Chollet on Deep Learning and the Meaning of Intelligence1 claim
- AMA | December 20201 claim
- AMA | July 20211 claim
- Why Do Mainframes Still Exist? What's Inside One? 40TB, 200+ Cores, AI, and more!1 claim
Most important claims · 9 of 9Search all claims for “Manifold” →
- Through training on massive data in high-dimensional space, LLMs build a smooth differentiable manifold that, due to compression pressure, encodes millions of reusable semantic 'programs'—vector functions mapping subsets of vector space to others—so the best way to think of an LLM is as a vast interpolative database of programs rather than a reasoning system.280 | François Chollet on Deep Learning and the Meaning of Intelligence
- Any pattern generated or found in nature can be efficiently discovered and modeled by a classical learning algorithm, because natural systems have structure imposed by evolutionary and selection processes, which creates a low-dimensional manifold a neural network can learn rather than requiring intractable brute-force enumeration.Demis Hassabis #2
- When we say the size of the universe we usually mean the observable universe, and the universe could be a closed positively-curved three-dimensional manifold with a finite size yet no middle, just as the two-dimensional surface of a sphere has no center because the 'middle' is inside and not part of the sphere itself.AMA | December 2020
- Z16 mainframes use closed-loop water cooling with deionized water, manifold distribution to each CPU socket, and hot-swappable CPUs that can be replaced without draining fluids from the systemWhy Do Mainframes Still Exist? What's Inside One? 40TB, 200+ Cores, AI, and more!
- The learnability principle extends beyond living things: mountains shaped by weathering, planetary orbits, and asteroid shapes have all survived repeated physical processes, so even non-biological natural systems carry a reverse-learnable pattern—'survival of the stablest'—whereas abstract man-made things like large-number factorization may have no such structure and require brute force.Demis Hassabis #2
- Although gene expression observations sit in a roughly 20,000-dimensional feature space, those observations lie on a very low-dimensional manifold within it, because not all locations in that space correspond to a system that can live; the goal is to discover the geometry of that manifold and use an observation's location on it to infer the system's behavior.251 | Rosemary Braun on Uncovering Patterns in Biological Complexity
- The 'rules of life' concept—the NSF's framing that Braun finds useful—refers to the constraints or organizing principles that map between the microscopic omics measurements and macroscopic observations like whether a cell is cancerous or an organism is a mouse or frog; discovering what dictates and constrains that microscopic-to-macroscopic relationship is the central problem, and asking why observations are constrained to a low-dimensional manifold is one instance of seeking such a rule.251 | Rosemary Braun on Uncovering Patterns in Biological Complexity
- Nonlinear dimension reduction and manifold learning use the local geometry of observations to stitch together the surface on which data lie, but a still-unanswered challenge is the question of scale: how large a local neighborhood to use, whether the same scale applies in every region, and how to tune that scale adaptively.251 | Rosemary Braun on Uncovering Patterns in Biological Complexity
- Physical space and distance may be emergent from the entanglement between abstract quantum degrees of freedom — not entanglement between macroscopic objects like people, but between invisible degrees of freedom in the space between them — where specific geometric patterns of entanglement knit together into a smooth spacetime manifold; however, whether these entanglement structures merely resemble distances or actually constitute them requires a dynamical theory linking entanglement to signal propagation speed, which is still unresolved.AMA | July 2021