2012/8/22 COSC 494/594 Unconventional Computation Introduction Fall 2012 Unconventional Computation 1 Course Information l l Instructor: Bruce MacLennan Course website: web.eecs.utk.edu/~mclennan/Classes/494-UC or 594-UC l Email: maclennan@eecs.utk.edu l Prereqs: linear algebra (basic CS, physics) l Grading: - Homework (every week or two) - Project or two - Term paper (on some kind of unconventional computation) - Presentation (for 594) Fall 2012 Unconventional Computation 2 1 2012/8/22 Course Outline I. Introduction II. Physics of computation III. Quantum computation IV. Molecular computation V. Analog computation? VI. Grad presentations on other unconventional computing paradigms Fall 2012 Unconventional Computation 3 Unconventional Computation l Unconventional (or non-standard) computation refers to the use of non-traditional technologies and computing paradigms - l Why would you want to do this? Hypercomputation or super-Turing computation refers to computation “beyond the Turing limit” - Fall 2012 Is this possible? Unconventional Computation 4 2 2012/8/22 Post-Moore’s Law Computation l The end of Moore’s Law is in sight! l Physical limits to: - density of binary logic devices - speed of operation l Requires a new approach to computation l Significant challenges l Will broaden & deepen concept of computation in natural & artificial systems Fall 2012 Unconventional Computation 5 Gate Energy Trends Trend of ITRS Min.'97-'03 Transistor Switching Energy Based on ITRS ’97-03 roadmaps 1.E-14 250 180 1.E-15 130 90 CVV/2 energy, J 1.E-16 LP min gate energy, aJ HP min gate energy, aJ 100 k(300 K) ln(2) k(300 K) 1 eV k(300 K) Node numbers (nm DRAM hp) 65 45 32 1.E-17 fJ 22 Practical limit for CMOS? 1.E-18 Na ïve lin ear ex Room-temperature 100 kT reliability limit One electron volt 1.E-19 1.E-20 1.E-21 aJ tra pol atio n Room-temperature kT thermal energy Room-temperature von Neumann - Landauer limit 1.E-22 1995 2000 2005 2010 2015 2020 zJ 2025 2030 2035 2040 2045 Year 2005/05/05 M. Frank, "Introduction to Reversible Computing" 6 3 2012/8/22 Differences in Spatial Scale 2.71828 0010111001100010100 … Fall 2012 … Unconventional Computation (Images from Wikipedia) 7 X := Y / Z Differences in Time Scale Fall 2012 P[0] := N i := 0 while i < n do if P[i] >= 0 then q[n-(i+1)] := 1 P[i+1] := 2*P[i] - D else q[n-(i+1)] := -1 P[i+1] := 2*P[i] + D end if i := i + 1 end while Unconventional Computation (Images from Wikipedia) 8 4 2012/8/22 Convergence of Scales Fall 2012 Unconventional Computation 9 Implications of Convergence l Computation on scale of physical processes l Fewer levels between computation & realization l Less time for implementation of operations l l Computation will be more like underlying physical processes Post-Moore’s Law computing greater assimilation of computation to physics Fall 2012 Unconventional Computation 10 5 2012/8/22 Computation is Physical “Computation is physical; it is necessarily embodied in a device whose behaviour is guided by the laws of physics and cannot be completely captured by a closed mathematical model. This fact of embodiment is becoming ever more apparent as we push the bounds of those physical laws.” — Susan Stepney (2004) Fall 2012 Unconventional Computation 11 Cartesian Duality in CS l Programs as idealized mathematical objects l Software treated independently of hardware l Focus on formal rather than material l Post-Moore’s Law computing: - less idealized - more dependent on physical realization l More difficult l But also presents opportunities… Fall 2012 Unconventional Computation 12 6 2012/8/22 Strengths of “Embodied Computation” l l l Information often implicit in: - its physical realization - its physical environment Many computations performed “for free” by physical substrate Representation & information processing emerge as regularities in dynamics of physical system Fall 2012 Unconventional Computation Example: Diffusion l l l l Fall 2012 13 Occurs naturally in many fluids Can be used for many computational tasks - broadcasting information - massively parallel search for optimization, constraint satisfaction etc. Expensive with conventional computation Free in many physical systems Unconventional Computation 14 7 2012/8/22 Example: Saturation l l l Sigmoids in ANNs & universal approx. Many physical systems have sigmoidal behavior - Growth process saturates - Resources become saturated or depleted Embodied computation uses free sigmoidal behavior Fall 2012 Unconventional Computation 15 (Images from Bar-Yam & Wikipedia) Example: Negative Feedback l l l Fall 2012 Positive feedback for growth & extension Negative feedback for: - stabilization - delimitation - separation - creation of structure Free from - evaporation - dispersion - degradation Unconventional Computation 16 8 2012/8/22 l l l Many algorithms use randomness - escape from local optima - symmetry breaking - deadlock avoidance - exploration Example: Randomness For free from: - noise - uncertainty - Imprecision “Free variability” Fall 2012 Unconventional Computation (Image from Anderson) 17 “Respect the Medium” l Conventional computer technology “tortures the medium” to implement computation l Embodied computation “respects the medium” l Goal of embodied computation: Exploit the physics, don’t circumvent it Fall 2012 Unconventional Computation 18 9 2012/8/22 But is it Computing? Fall 2012 Unconventional Computation 19 Some Non-Turing Characteristics of Embodied Computation l Operates in real time and real space - with real matter and real energy - and hence non-ideal aspects of physical realization l Often does not terminate l Often has no distinct inputs or outputs l Often purpose is not to get an answer from an input l Often purpose is not to control fixed agent l Different notions of equivalence and universality Fall 2012 Unconventional Computation 20 10 2012/8/22 Is EC a Species of Computing? l l l l The Turing Machine provides a precise definition of computation Embodied computation may seem imprecise & difficult to discriminate from other physical processes Expanding concept of computation beyond TM requires an expanded definition Fall 2012 Unconventional Computation 21 Non-Turing Computation Fall 2012 Unconventional Computation 22 11 2012/8/22 Frames of Relevance l l l l CT computation is a model of computation All models have an associated frame of relevance - determined by model’s simplifying assumptions - by aspects & degrees to which model is similar to modelled system Determine questions model is suited to answer Using outside FoR may reflect model & simplifying assumptions more than modelled system Fall 2012 Unconventional Computation 23 Models & Simplifying Assumptions l l l l l l Fall 2012 Turing computation is a model of computation A model is like its subject in relevant ways Unlike it in irrelevant ways A model is suited to pose & answer certain classes of questions Thus every model exists in a frame of relevance (FoR) FoR defines domain of reliable use of model Unconventional Computation 24 12 2012/8/22 Example: FoR of Maps 2010/03/22 25 The FoR of Turing Computation l Historical roots: issues of formal calculability & provability in axiomatic mathematics; hence: - finite number of steps & finite but unlimited resources - computation viewed as function evaluation - discreteness assumptions Fall 2012 Unconventional Computation 26 13 2012/8/22 Idealizing Assumptions l Finite but unbounded resources l Discreteness & definiteness l Sequential time l l Computational task = evaluation of well-defined function Computational power defined in terms of sets of functions Fall 2012 Unconventional Computation 27 Alternate Frames of Relevance for Expanded Notions of Computation l l l Fall 2012 Natural Computation - applying natural processes in computation - alternative realizations of formal processes Nanocomputation - direct realizations of non-Turing computations - unique characteristics Quantum & Quantum-like Computation Unconventional Computation 28 14 2012/8/22 Natural Computation l l l l Natural computation = computation occurring in nature or inspired by it Occurs in nervous systems, DNA, microorganisms, animal groups Good models for robust, efficient & effective artificial systems (autonomous robots etc.) Different issues are relevant Fall 2012 Unconventional Computation 29 Relevant Issues Outside TC FoR l Real-time control l Continuous computation l Robustness l Generality, flexibility & adaptability l Non-functional computation Fall 2012 Unconventional Computation 30 15 2012/8/22 Relevant Issues Outside TC FoR l Error, noise & uncertainty are unavoidable must be part of model of computation - may be used productively - l Microscopic reversibility may occur e.g., reversible chemical reactions - want statistical or macroscopic progress - l Computation proceeds asynchronously in continuous-time parallelism Fall 2012 Unconventional Computation 31 Real-Time Control l Real-time (RT) response constraints l Asymptotic complexity is usually irrelevant l - Input size typically constant or of limited variability - Computational resources are bounded Relevant: relation of RT response rate to RT rates of its components Fall 2012 Unconventional Computation 32 16 2012/8/22 Continuous Computation l Inputs & outputs often: - Are continuous quantities - Vary continuously in real time l Computational processes often continuous l More or less powerful than TMs? l Obviously can be approximated by discrete quantities varying at discrete times, but … Fall 2012 Unconventional Computation 33 “Metaphysics” of Reals l l l “Metaphysical issues”: - Turing-computable reals vs. standard reals - Standard reals vs. non-standard reals Results depend on “metaphysical issues” ⇒ outside FoR of model Naïve real analysis is sufficient for models of natural computation Fall 2012 Unconventional Computation 34 17 2012/8/22 Cross-Frame Comparisons l l l l l Fall 2012 Can we compare models with different FoRs? Yes: can translate one to other’s FoR Typically make incompatible simplifying assumptions Results may depend on specifics of translation E.g., how are continuous quantities represented in TC? Unconventional Computation 35 Within-Frame Comparison Comparison OK Model 1 Model 2 Common Frame of Relevance Fall 2012 Unconventional Computation 36 18 2012/8/22 Cross-Frame Comparison Meaningless Comparison Model 2 Model 1 FoR A FoR B Fall 2012 Unconventional Computation 37 Translated Comparison Meaningful Comparison, But Relevant? translation Model 1 FoR A Fall 2012 Model 1ʹ′ Model 2 FoR B Unconventional Computation 38 19 2012/8/22 Translation to Third Frame Relevant? translation Model 1 FoR A Fall 2012 translation Model 1ʹ′ Model 2ʹ′ FoR C Model 2 FoR B Unconventional Computation 39 Super-Turing vs. Non-Turing l l Notion of Super-Turing computation is relative to FoR of Turing computation Super-Turing computation is important, but so is Non-Turing computation Fall 2012 Unconventional Computation 40 20 2012/8/22 Some Issues in Non-Turing Computation l l l l l Fall 2012 What is computation in broad sense? What FoRs are appropriate for non-Turing computation? Models of non-Turing computation How fundamentally to incorporate error, uncertainty, imperfection, reversibility? How systematically to exploit new physical processes? Unconventional Computation 41 Expanding the Range of Physical Computation l Digital VLSI becoming a vicious cycle? l A limit to the number of bits and MIPS l l l Alternative technologies are surpassed before they can be developed False assumption that binary logic is the only way to compute How to break out of the vicious cycle? Fall 2012 Unconventional Computation 42 21 2012/8/22 What is Computation? l l l l What distinguishes computing (physically realized information processing) from other physical processes? Computation is a mechanistic process, the purpose or function of which is the abstract manipulation (processing) of abstract objects Purpose is formal rather than material Does not exclude embodied computation, which relies more on physical processes Fall 2012 Unconventional Computation 43 Possible Physical Realizations of Computation l Any abstract manipulation of abstract objects is a potential computation de novo applications of math models - applications suggested by natural computation - l l But it must be physically realizable Any reasonably controllable, mathematically described, physical process can be used for computation Fall 2012 Unconventional Computation 44 22 2012/8/22 Some Requirements l l l Speed, but: - faster is not always better - slower processes may have other advantages Feasibility of required transducers Accuracy, stability & controllability as required for the application - Fall 2012 natural computation shows ways of achieving, even with imperfect components Unconventional Computation 45 Matching Computational & Physical Processes l l Familiarity of binary logic maintains vicious cycle Natural computation shows alternate modes of computation, e.g.: information processing & control in brain - emergent self-organization in animal societies - l l Fall 2012 Openness to usable physical processes Library of well-matched computational methods & physical realizations Unconventional Computation 46 23 2012/8/22 General-Purpose Computation l l l l Value of general-purpose computers for all modes of computation “Universality” is relative to frame of relevance E.g., speed of emulation is essential to realtime applications (natural computation) Merely computing the same function may be irrelevant Fall 2012 Unconventional Computation 47 Conclusions l Turing model of computation exists in a frame of relevance - not appropriate to natural computation, nanocomputation, quantum / quantum-like computation - central issues of these include continuity, indeterminacy, parallelism l Broader definition of computation is needed l Facilitates new implementation technologies l Fall 2012 Improves understanding of computation in nature Unconventional Computation 48 24