Three centuries of discovery

Open the full timeline as a list

1665 — Huygens: two pendulum clocks on one beam fall into step — synchrony, first noticed [INST·23 The Chorus]
1687 — Newton: universal gravitation — and the two-body ellipse, solved exactly [INST·17 The Orbit]
1733 — de Moivre: the first normal curve, as the limit of many coin flips [INST·19 The Walk]
1747 — d'Alembert: solves the vibrating-string wave equation [INST·15 The String]
1760 — Bernoulli: the first mathematical epidemic model — weighing smallpox inoculation against the disease [INST·28 The Contagion]
1763 — Bayes: belief updates as posterior ∝ likelihood × prior — the rule the filter runs [INST·25 The Lens]
1772 — Lagrange: the equilateral three-body solution and the L-points [INST·17 The Orbit]
1787 — Chladni: bows a sand-strewn metal plate and the grains leap off the moving regions to settle on the still nodal lines — the first images of a vibration’s modes [INST·47 The Drum]
1801 — Young: the original two-slit experiment — light interferes, so light is a wave [INST·11 The Slit]
1809 — Gauss: least squares to track Ceres from a few noisy sightings — estimation is born [INST·25 The Lens]
1810 — Laplace: the Central Limit Theorem in its general form [INST·19 The Walk]
1822 — Fourier: claims any periodic signal is a sum of sines (Théorie analytique de la chaleur) [INST·01 The Spectrum] [INST·15 The String]
1822 — Navier: writes the momentum equation — viscosity and all [INST·03 The Flow]
1834 — Scott Russell: observes a solitary wave of translation on the Union Canal near Edinburgh — a heap of water that travels a mile without changing shape [INST·40 The Soliton]
1845 — Stokes: puts it on rigorous footing: the modern Navier–Stokes form [INST·03 The Flow]
1847 — Cauchy: method of steepest descent — follow the negative gradient downhill, the optimiser still in use [INST·27 The Descent]
1848 — Wilbraham: first spots the overshoot at a jump — then it is forgotten for fifty years [INST·01 The Spectrum]
1865 — Clausius: names entropy and states the second law of thermodynamics [INST·18 The Arrow]
1867 — Maxwell: proposes a tiny "demon" that sorts fast from slow molecules — apparently defeating the second law without doing work [INST·39 The Demon]
1872 — Boltzmann: the H-theorem and S = k ln Ω — irreversibility from counting [INST·18 The Arrow]
1876 — Loschmidt: the reversibility objection: symmetric laws, an asymmetric world [INST·18 The Arrow]
1883 — Reynolds: his number, and the dye experiment that catches laminar flow breaking up [INST·03 The Flow]
1887 — Michelson–Morley: find no aether — light's speed refuses to add up [INST·22 The Cone]
1890 — Poincaré: no closed form for three bodies — the first sight of chaos [INST·05 The Divergence] [INST·17 The Orbit]
1890 — Poincaré: recurrence: a closed system must eventually return near its start [INST·18 The Arrow]
1892 — Lyapunov: his stability theory gives the exponent λ that measures the stretch [INST·05 The Divergence]
1895 — Korteweg & de Vries: derive the KdV equation for shallow water waves and find its exact soliton solution, vindicating Scott Russell 60 years later [INST·40 The Soliton]
1899 — Gibbs: explains it: the ripple narrows but homes in on ≈8.95%, never vanishing [INST·01 The Spectrum]
1900 — Bachelier: models market prices as a random walk — finance meets diffusion, the first time-series [INST·19 The Walk] [INST·24 The Oracle]
1905 — Einstein: Brownian motion ties the walk to atoms and Avogadro's number [INST·19 The Walk] [INST·43 The Bloom]
1905 — Einstein: special relativity: c is invariant, and space and time mix [INST·22 The Cone]
1906 — Markov: the Markov chain — a process whose next step depends only on the present, with a long-run equilibrium [INST·32 The Rank] [INST·45 The Regime]
1908 — Minkowski: spacetime — the geometry in which the light cone is absolute [INST·22 The Cone]
1908 — Langevin: the Langevin equation — Newton plus a random force; the stochastic differential equation that underlies every diffusion here [INST·43 The Bloom]
1909 — Haar: the first wavelet — a square step that splits a signal by scale [INST·37 The Loom]
1911 — Weyl: Weyl’s law — the count of eigenfrequencies below a bound grows with the drum’s area, so the spectrum does betray the size, if not the shape [INST·47 The Drum]
1912 — Perron & Frobenius: a positive matrix has one largest eigenvalue with an all-positive eigenvector — the unique, well-defined rank [INST·32 The Rank]
1915 — Einstein: general relativity: matter curves spacetime, and light follows the curve [INST·07 The Well] [INST·48 The Horizon]
1915 — Whittaker: the interpolation formula that rebuilds a signal from its samples [INST·08 The Strobe]
1916 — Schwarzschild: the first exact solution — and with it, the event horizon [INST·07 The Well] [INST·48 The Horizon]
1918 — Julia & Fatou: the dynamics of z² + c — decades before anyone could see it [INST·21 The Set]
1919 — Eddington: the eclipse measures the bending at twice Newton's value — Einstein, famous overnight [INST·07 The Well]
1924 — Bose: new statistics for light quanta — particles that bunch together [INST·14 The Condensate]
1924 — de Broglie: matter has a wavelength too — so particles can interfere [INST·11 The Slit]
1925 — Einstein: extends Bose statistics to atoms and predicts the condensate [INST·14 The Condensate]
1925 — Ising: solves the 1D chain — and finds no phase transition (the model nearly dies here) [INST·04 The Threshold] [INST·36 The Engram]
1925 — Lotka: derives the oscillating predator–prey equations in Elements of Physical Biology — a population read as a chemical-style dynamical system [INST·44 The Tide]
1926 — Schrödinger: the wave equation — the whole future of ψ [INST·12 The Wavefunction]
1926 — Born: |ψ|² is probability: the cloud itself (Nobel 1954) [INST·12 The Wavefunction]
1926 — Volterra: explains why the WWI pause in Adriatic fishing raised the shark fraction — the same equations, from his son-in-law fish-market data [INST·44 The Tide]
1927 — Hund: first spots tunnelling, in the splitting of molecular spectra [INST·13 The Tunnel]
1927 — Madelung: rewrites ψ as a fluid — a density and a flowing phase [INST·12 The Wavefunction]
1927 — Kermack & McKendrick: the SIR model and the epidemic threshold: an outbreak needs R₀ > 1 to take off [INST·28 The Contagion]
1928 — Gamow: explains alpha decay by tunnelling — radioactivity as a quantum leak [INST·13 The Tunnel]
1928 — Nyquist: fixes the rate: sample above twice the top frequency [INST·08 The Strobe]
1929 — Hubble: galaxies recede at a speed proportional to their distance — the universe is expanding [INST·50 The Web]
1929 — Szilárd: resolves the demon with a one-molecule engine: one measurement yields kT ln 2 of work — and costs one bit of memory [INST·39 The Demon]
1933 — Kotelnikov: proves the sampling theorem in full [INST·08 The Strobe]
1933 — Zwicky: the Coma cluster moves far too fast for its visible mass — most of the gravitating matter is unseen [INST·50 The Web]
1934 — Gause: tests predator–prey in test tubes of Paramecium and Didinium — the predator eats all prey then starves; coexistence needs a refuge [INST·44 The Tide]
1935 — Einstein, Podolsky & Rosen: argue QM must be incomplete — "spooky action at a distance" can't be real [INST·38 The Pact]
1936 — Whitney: the embedding theorem — any smooth d-manifold sits faithfully inside ℝ^(2d+1) [INST·31 The Shadow]
1937 — Landau: the general theory of phase transitions and the order parameter [INST·04 The Threshold]
1938 — London: ties superfluid helium to Bose–Einstein condensation [INST·14 The Condensate]
1944 — Onsager: solves the 2D Ising model exactly — T_c and the critical exponents, by hand [INST·04 The Threshold]
1944 — Gutenberg & Richter: earthquake magnitudes follow a power law — no characteristic size, decades before anyone could say why [INST·29 The Sandpile]
1944 — Itô: stochastic calculus: the lemma that integrates dS exactly into a log-normal [INST·24 The Oracle]
1944 — von Neumann & Morgenstern: Theory of Games and Economic Behavior founds game theory — rational players, payoff matrices and the minimax theorem [INST·33 The Tournament]
1946 — Gabor: tiles time–frequency into "logons" and finds the joint-resolution limit [INST·37 The Loom]
1948 — Shannon: 'A Mathematical Theory of Communication' — entropy, capacity, the whole field [INST·20 The Code]
1948 — von Neumann & Ulam: invent cellular automata — a self-reproducing machine on a grid [INST·09 The Rule] [INST·35 The Garden]
1949 — Onsager: circulation is quantised: vortices carry integer charge [INST·12 The Wavefunction]
1949 — Shannon: makes the sampling theorem a cornerstone of digital signal processing [INST·08 The Strobe]
1949 — Ulam · Metropolis: the Monte Carlo method — answer by sampling when the formula is hard [INST·24 The Oracle] [INST·34 The Anneal]
1949 — Hebb: 'neurons that fire together wire together' — memory as the strengthening of a connection [INST·36 The Engram]
1950 — Hamming: the first error-correcting code — parity that names the flipped bit [INST·20 The Code]
1950 — Nash: the Nash equilibrium — every finite game has a stable strategy profile no player can beat by deviating alone [INST·33 The Tournament]
1951 — Belousov: discovers a chemical reaction that oscillates in colour instead of settling to equilibrium — rejected as impossible by the journals, and ignored for a decade [INST·46 The Spiral]
1952 — Turing: 'The Chemical Basis of Morphogenesis' — diffusion can make pattern, not just smooth it [INST·06 The Skin]
1953 — Metropolis, Rosenbluth & Teller: the Metropolis algorithm — sample a Boltzmann distribution by accepting moves with probability min(1, e^−ΔE/T) [INST·34 The Anneal]
1957 — Esaki: the tunnel diode puts the effect to work (Nobel 1973) [INST·13 The Tunnel]
1957 — Broadbent & Hammersley: founding paper on percolation — originally motivated by fluid flow through random media [INST·26 The Percolation]
1957 — Bellman: dynamic programming and the principle of optimality — the value of a state is its best reward plus the discounted value of where you land; also names the curse of dimensionality [INST·42 The Maze]
1958 — Rosenblatt: the perceptron: a learning machine of weighted neurons — but a single layer cannot solve XOR [INST·27 The Descent]
1959 — Holling: the functional response — a predator saturates because prey take time to handle (type II); the realism that lets the model cycle and crash [INST·44 The Tide]
1960 — Reed & Solomon: codes that armour CDs, QR codes and deep-space links [INST·20 The Code]
1960 — Kalman: the recursive filter: optimal estimation of a linear system, one step at a time [INST·25 The Lens]
1961 — Jönsson: runs the double slit with single electrons, one at a time [INST·11 The Slit]
1961 — Landauer: shows that erasing one bit of information dissipates at least kT ln 2 as heat — information is physical [INST·39 The Demon]
1961 — FitzHugh & Nagumo: reduce the Hodgkin–Huxley nerve impulse to two variables — a fast excitable voltage and a slow recovery — the minimal model of a firing, refractory medium [INST·46 The Spiral]
1963 — Lorenz: three equations for convection — the butterfly effect, chaos found by accident from a rounded printout [INST·05 The Divergence] [INST·16 The Attractor] [INST·31 The Shadow]
1963 — Rosenzweig & MacArthur: a graphical predator–prey model with logistic prey and saturating predation — isoclines that predict a stable point or a limit cycle [INST·44 The Tide]
1964 — Sharkovskii: orders the periods: a map with a 3-cycle must have cycles of every length [INST·02 The Cascade]
1964 — Bell: turns "is the world local?" into an inequality you can actually measure [INST·38 The Pact]
1964 — Zhabotinsky: revives Belousov’s reaction and finds it makes travelling and rotating waves in a dish — chemistry that propagates like a nerve impulse or a heartbeat [INST·46 The Spiral]
1965 — Penrose: collapse to a singularity is inevitable — black holes are a robust prediction of relativity, not a fluke (Nobel 2020) [INST·48 The Horizon]
1965 — Cooley & Tukey: the FFT makes the transform cheap — now inside MP3, JPEG, every analyser [INST·01 The Spectrum]
1965 — Kirkpatrick et al.: Monte Carlo computation of p_c for the square lattice ≈ 0.5927 [INST·26 The Percolation]
1965 — Samuelson: geometric (not arithmetic) Brownian motion for prices — they can never go negative [INST·24 The Oracle]
1965 — Zabusky & Kruskal: rediscover solitons on a computer, observe that two pulses pass through each other unchanged — and coin the name "soliton" [INST·40 The Soliton]
1966 — Kac: “Can one hear the shape of a drum?” — asks whether two differently-shaped membranes could ever share an identical spectrum of tones [INST·47 The Drum]
1967 — Winfree: biological oscillators sync once the coupling beats the spread [INST·23 The Chorus]
1967 — Mandelbrot: 'How Long Is the Coast of Britain?' — fractional dimension enters science: the length depends on the ruler [INST·30 The Dendrite]
1967 — Viterbi: the Viterbi algorithm — dynamic programming recovers the single most-likely hidden path through a noisy sequence, exactly, in one backward sweep [INST·45 The Regime]
1968 — Veneziano: his amplitude for the strong force secretly describes a string [INST·15 The String]
1969 — Clauser, Horne, Shimony & Holt: the experimentally testable form of Bell's bound, S ≤ 2 [INST·38 The Pact]
1969 — Apollo: the Kalman filter flies the guidance computer to the Moon and back [INST·25 The Lens]
1970 — Nambu, Nielsen & Susskind: read it literally: the constituents are vibrating strings [INST·15 The String]
1970 — Conway: the Game of Life makes cellular automata famous [INST·09 The Rule] [INST·35 The Garden]
1970 — Gosper: finds the glider gun — Life grows without bound, so it can store and move information [INST·35 The Garden]
1970 — Zel'dovich: the approximation that lets gravity be run by hand — matter collapses first into sheets (pancakes), then filaments, then knots [INST·50 The Web]
1970 — Baum, Petrie, Soules & Weiss: the hidden Markov model with forward–backward and Baum–Welch — infer the hidden states, and the chain itself, from the observations alone [INST·45 The Regime]
1971 — Wilson: the renormalisation group explains universality (Nobel 1982) [INST·04 The Threshold]
1971 — Hafele–Keating: flying atomic clocks confirm time dilation directly [INST·22 The Cone]
1971 — Rosenzweig: the paradox of enrichment — raising the prey carrying capacity destabilises the equilibrium into ever-wider cycles that can drive extinction [INST·44 The Tide]
1972 — Gierer & Meinhardt: name the design rule: short-range activation, long-range inhibition [INST·06 The Skin]
1972 — May: asks whether a large complex system is stable and finds it usually is not — richly connected ecosystems tend to be less stable, not more [INST·44 The Tide]
1972 — Winfree: shows the rotating chemical spiral turns about a phase singularity — a point where phase is undefined — and ties the same geometry to cardiac fibrillation [INST·46 The Spiral]
1973 — Black · Scholes · Merton: the same log-normal, priced: option value as a discounted forecast [INST·24 The Oracle]
1973 — Maynard Smith & Price: the evolutionarily stable strategy — game theory carried into biology, where fitness rather than reason selects the move [INST·33 The Tournament]
1974 — Little: casts a neural network as a spin system — persistent firing states are the stored memories [INST·36 The Engram]
1975 — Li & Yorke: coin the word — “Period Three Implies Chaos” [INST·02 The Cascade]
1975 — Kuramoto: the exactly solvable synchronization model — order parameter r and the critical K_c [INST·23 The Chorus]
1976 — Robert May: the logistic map as a toy ecology (Nature) — a simple law, an uncomputable fate [INST·02 The Cascade]
1978 — Feigenbaum: finds the doubling ratio δ ≈ 4.6692 is universal — on a pocket calculator [INST·02 The Cascade]
1979 — Kaplan–Yorke: conjecture links the Lyapunov exponents to the attractor's fractal dimension [INST·16 The Attractor]
1980 — Benettin: an algorithm to measure λ from a shadow orbit — the method the rack uses live [INST·05 The Divergence] [INST·16 The Attractor]
1980 — Tsirelson: proves quantum mechanics can violate CHSH only up to 2√2 [INST·38 The Pact]
1980 — Mandelbrot: computes the set at IBM and reveals the shape [INST·21 The Set] [INST·49 The Bulb]
1980 — Peebles: the gravitational theory of how faint early-universe ripples grow into the cosmic web (Nobel 2019) [INST·50 The Web]
1980 — Packard, Crutchfield, Farmer & Shaw: “Geometry from a Time Series” — reconstruct an attractor from a single signal’s own delayed copies [INST·31 The Shadow]
1981 — Binnig & Rohrer: the scanning tunnelling microscope feels single atoms (Nobel 1986) [INST·13 The Tunnel]
1981 — Witten & Sander: diffusion-limited aggregation — random walkers stick on contact into a branching fractal of dimension ≈ 1.71 [INST·30 The Dendrite]
1981 — Takens: proves it: delay-coordinate embedding is generically diffeomorphic to the true attractor — same invariants [INST·31 The Shadow]
1981 — Axelrod & Hamilton: the iterated tournament — Tit-for-Tat beats every cleverer rule; cooperation evolves among selfish agents that meet again [INST·33 The Tournament]
1981 — Benzi, Sutera & Vulpiani: introduce stochastic resonance to explain the 100,000-year ice-age cycle: orbital forcing alone is too weak — noise from climate fluctuations makes it detectable [INST·41 The Noise]
1982 — Sparrow: the definitive study of the Lorenz equations [INST·16 The Attractor]
1982 — Aspect: switches the analysers in flight — closes the locality loophole, S ≈ 2.7 [INST·38 The Pact]
1982 — Nienhuis: exact critical exponents β = 5/36, ν = 4/3 via Coulomb gas / conformal field theory [INST·26 The Percolation]
1982 — Hopfield: a recurrent net whose energy only falls, so memories are the basins of attraction it rolls into [INST·36 The Engram]
1982 — Berlekamp, Conway & Guy: 'Winning Ways' proves Life is Turing-complete — a universal computer built from one rule [INST·35 The Garden]
1982 — Douady & Hubbard: prove M is connected, and give it his name [INST·21 The Set]
1982 — Bennett: proves the demon cannot violate the second law: the erasure of its memory ledger, not the measurement, is the thermodynamic cost [INST·39 The Demon]
1982 — Anderson: the reverse-time SDE — a diffusion run backward is itself a diffusion, driven by the score of the density; the theorem generative models would later stand on [INST·43 The Bloom]
1983 — Gray & Scott: the autocatalytic reaction–diffusion model this engine runs [INST·06 The Skin]
1983 — Wolfram: sorts the 256 elementary rules into four behaviours [INST·09 The Rule] [INST·35 The Garden]
1983 — Grassberger & Procaccia: the correlation dimension — read a strange attractor’s fractal dimension straight off the reconstructed cloud [INST·31 The Shadow]
1983 — Kirkpatrick, Gelatt & Vecchi: Optimization by Simulated Annealing — melt a hard problem, then cool it slowly through Metropolis moves into its best state [INST·34 The Anneal]
1984 — Morlet & Grossmann: formalise the continuous wavelet transform, out of seismic traces [INST·37 The Loom]
1984 — Brady & Ball: copper electrodeposits grow as DLA fractals — the model caught in the lab, the same dimension measured in metal [INST·30 The Dendrite]
1985 — Amit, Gutfreund & Sompolinsky: the spin-glass calculation: the net stores ≈0.138N patterns before the basins dissolve [INST·36 The Engram]
1985 — Černý: independently applies thermodynamic annealing to the travelling-salesman problem — the method arrives twice at once [INST·34 The Anneal]
1986 — Rumelhart, Hinton & Williams: backpropagation trains hidden layers efficiently — the multilayer network finally learns [INST·27 The Descent]
1986 — Reynolds: Boids: three local rules — align, cohere, separate [INST·10 The Flock]
1987 — Bak, Tang & Wiesenfeld: self-organized criticality: a sandpile tunes itself to the critical point and emits 1/f noise — no parameter set by hand [INST·29 The Sandpile]
1988 — Daubechies: compactly-supported orthonormal wavelets — the basis inside JPEG 2000 [INST·37 The Loom]
1988 — Sutton: temporal-difference learning — update a prediction from a later, better prediction, with no model of the world; the bracket that becomes Q-learning [INST·42 The Maze]
1989 — Laskar: the inner Solar System is chaotic, with a ~5-million-year horizon [INST·17 The Orbit]
1989 — Mallat: multiresolution analysis + the fast wavelet transform [INST·37 The Loom]
1989 — Hart, Sandin & Kauffman: distance-estimated ray tracing — march a ray by a safe lower bound on the distance to the surface, which is what makes a 3D escape-time fractal renderable at all [INST·49 The Bulb]
1989 — Watkins: Q-learning — a model-free, off-policy rule that provably converges to the optimal action-values just by trying, watching and updating [INST·42 The Maze]
1989 — Hamilton: the Markov-switching model carries hidden-state inference into economics — reading booms and recessions as latent regimes behind the noise of GDP and returns [INST·45 The Regime]
1990 — De Kepper: first chemical Turing patterns seen in the lab (the CIMA reaction) [INST·06 The Skin]
1990 — Dhar: the abelian sandpile: topplings commute, the recurrent states form a group, and the identity is a fractal [INST·29 The Sandpile]
1991 — Shishikura: proves the Mandelbrot boundary has Hausdorff dimension 2 [INST·21 The Set]
1992 — Nowak & May: spatial games — cooperators and defectors on a grid form ever-shifting fractal patterns; structure alone sustains cooperation [INST·33 The Tournament]
1992 — Tesauro: TD-Gammon reaches near-expert backgammon by playing millions of games against itself — temporal-difference learning, scaled up, beats hand-tuned programs [INST·42 The Maze]
1992 — Gordon, Webb & Wolpert: answers Kac — no: they build two different polygonal drums with exactly the same spectrum. You can hear the area and perimeter, but not the shape [INST·47 The Drum]
1993 — Berrou: turbo codes reach within a whisker of the Shannon limit [INST·20 The Code]
1994 — Watts & Strogatz · Barabási: percolation transitions in real-world networks: internet, epidemics, markets [INST·26 The Percolation] [INST·32 The Rank]
1994 — Moss, Pierson & O'Gorman: demonstrate stochastic resonance in biological neurons — sensory systems appear to operate near the optimal noise level for signal detection [INST·41 The Noise]
1995 — Englert–Greenberger–Yasin: pin down the exact duality relation V² + D² ≤ 1 [INST·11 The Slit]
1995 — Cornell, Wieman & Ketterle: the first gaseous BEC, at JILA and MIT (Nobel 2001) [INST·14 The Condensate]
1995 — Vicsek: the minimal model — collective motion as a true phase transition [INST·10 The Flock]
1998 — Watts & Strogatz: small-world networks — a few long-range links collapse the distance a contagion must travel [INST·28 The Contagion]
1998 — Page & Brin: PageRank — rank the web by its link structure alone; a random surfer’s equilibrium, and the founding of Google [INST·32 The Rank]
1998 — Gammaitoni, Hänggi, Jung & Marchesoni: publish the landmark review of stochastic resonance, establishing it as a universal phenomenon in nonlinear noisy systems [INST·41 The Noise]
1999 — Jos Stam: “Stable Fluids” — an unconditionally stable solver; real-time fluids in a browser [INST·03 The Flow]
2001 — Pastor-Satorras & Vespignani: on scale-free networks the epidemic threshold can vanish — hubs change everything [INST·28 The Contagion]
2002 — Couzin: maps the switch between swarm, torus and aligned flock [INST·10 The Flock]
2004 — Cook: proves Rule 110 is Turing-complete — a universal computer [INST·09 The Rule]
2005 — Strogatz: the Millennium Bridge wobble: pedestrians lock step with the sway [INST·23 The Chorus]
2008 — Ballerini: real murmurations: each bird tracks ~7 neighbours, by rank not distance [INST·10 The Flock]
2009 — White & Nylander: the Mandelbulb — take the nth power of a point in spherical coordinates and z → zⁿ + c becomes a 3D fractal with genuine surface detail [INST·49 The Bulb]
2012 — Krizhevsky, Sutskever & Hinton: AlexNet: the same gradient descent, scaled to GPUs, wins ImageNet and ignites deep learning [INST·27 The Descent]
2012 — Sugihara et al.: convergent cross mapping — “shadow manifolds” turn reconstruction into a test for causation between time series [INST·31 The Shadow]
2015 — DeepMind: the Deep Q-Network learns to play 49 Atari games from raw pixels and the score alone, at human level — Q-learning with a deep net as the value function [INST·42 The Maze]
2015 — Sohl-Dickstein et al.: deep unsupervised learning by nonequilibrium thermodynamics — destroy data with a slow diffusion, then train a network to reverse each step [INST·43 The Bloom]
2016 — DeepMind: AlphaGo defeats Lee Sedol at Go — self-play reinforcement learning plus tree search solves a game long thought a decade away [INST·42 The Maze]
2019 — Event Horizon Telescope: photographs a black hole's shadow for the first time [INST·07 The Well] [INST·48 The Horizon]
2019 — Song & Ermon: score-based generative modelling — learn the gradient of the log-density and sample by following it with Langevin dynamics [INST·43 The Bloom]
2020 — Ho, Jain & Abbeel: denoising diffusion probabilistic models — the clean recipe that made diffusion the engine behind modern image generators [INST·43 The Bloom]
2021 — Song et al.: unifies diffusion as a reverse-time SDE with a deterministic probability-flow ODE twin — one framework for every sampler [INST·43 The Bloom]
2022 — Aspect, Clauser & Zeilinger: the Nobel for the experiments proving the world is not locally real [INST·38 The Pact]
2024 — Hopfield & Hinton: the Nobel Prize in Physics for the physics of memory and learning in neural networks [INST·36 The Engram]

TheIceJi