AI summary by TwinMind

Probabilistic Computing: A New Era?

Sabine Hossenfelder

10m 57s

Introduction to Randomness in Computing

  • Three distinct ways randomness improves computation:
    • Noisy computation
    • Stochastic computation
    • Probabilistic computation
  • Random numbers used extensively in financial management, climate models, AI
  • Computers not designed for randomness, typically use pseudo-random number generators (PRNGs)
    • PRNGs are deterministic algorithms producing seemingly random numbers
    • Example: digits of pi seeded from exact time of day
    • Consumes computational power unnecessarily given abundant natural randomness

Noisy Computation (True Random Number Generators)

  • Hardware random number generators exploit natural noise sources
    • Quantum sources: electronic noise
    • Non-quantum sources: chaos, metastable systems
  • In use ~10 years, primarily for cryptographic protocols
  • New research exploring noise effects on circuit operation:
    • Low noise: predictable behavior
    • Intermediate noise: random behavior
    • High noise: different predictable behavior (amplifies some processes, drowns others)
  • Noise level adjustment allows circuits to switch operational modes

Stochastic Computing

  • Different calculation method using bit strings
  • Traditional computing: bit strings represent numbers
  • Stochastic computing: bit strings represent probabilities
    • Only counts zeros/ones, position irrelevant
    • Example: "0100" = 25% true, 75% false
  • Multiplication property: bit-by-bit multiplication of probability strings yields correct result
    • Example calculation: 50% × 50% = 25%
    • Works for any string length and probability
  • Advantages: fast, easy, energy-efficient multiplication
  • Disadvantages:
    • String length determines accuracy
    • High accuracy requires long strings + extensive sampling
    • Increases energy consumption again
  • Historical context: idea from 1940s, barely used today
  • Current applications: Internet of Things devices where high accuracy unnecessary
    • Example: fridge knowing milk shortage with 99% probability sufficient

Probabilistic Computation

  • Uses specialized hardware for optimization problems
  • Analogy: sand on vibrating plate finds standing wave modes
    • Sound creates random motion
    • Sand accumulates where plate moves least (optimal positions)
  • System configured so optimal state answers the question
  • Applications: task scheduling, route optimization, fleet management, airplane boarding, supply chain, package delivery, taxi timing, Sudoku solving
  • Ising model most common implementation:
    • Individual states have spins
    • Spins couple according to specifications
    • System relaxes to ground state after randomness applied

Quantum vs Classical Annealing

  • D-Wave uses quantum annealing with entanglement between spins
    • Requires cooling to near absolute zero due to fragile entanglement
    • More powerful due to entanglement but expensive cooling requirement
  • Classical annealing doesn't need quantum effects
    • Uses true random number generators for randomness
    • Operates at room temperature
    • Extensive recent research development

Performance Comparison: Tokyo University Chip

  • Prototype specifications: 496 coupled spins
  • Benchmark test: maxcut problem (optimize network cuts)
  • Performance vs conventional computer:
    • Annealing chip: 2.9 watts, ~1,000 seconds
    • Standard computer: 200 watts, almost 1 hour
  • Other companies developing similar chips: Chiba, Fujitsu

Limitations and Challenges

  • Annealing chips don't scale easily (can't combine multiple units)
  • Noise level critical for optimization:
    • Too low: no optimization occurs
    • Too high: never finds optimal state
  • Quantum computing noise problem: researchers don't understand their system's noise behavior

Industry Trend Analysis

  • Movement from all-purpose to specialized hardware architecture
  • Comparison to Industrial Revolution coming to integrated circuits
  • Humorous prediction: integrated circuits forming labor unions

Brilliant.org Sponsorship

  • Educational platform for coding, math, science, problem-solving
  • Topics: solar panels, neural networks, astrophysics, special relativity, computational biology
  • Monthly new course additions
  • Special offer: brilliant.org/Sabina for 30-day free trial + 20% off annual premium

TwinMind

Get summaries like this for your own meetings

TwinMind takes notes in the background during meetings, lectures, and calls, then writes the summary for you.

Free plan to get started

No bot joins your calls

Phone, desktop & Chrome apps

Your next meeting, summarized like this

TwinMind listens in the background and turns meetings, lectures, and calls into searchable notes you can chat with. Start free.

Available on phone, desktop & Chrome

Your next meeting, summarized like this

TwinMind listens in the background and turns meetings, lectures, and calls into searchable notes you can chat with. Start free.

Available on phone, desktop & Chrome

Your next meeting, summarized like this

TwinMind listens in the background and turns meetings, lectures, and calls into searchable notes you can chat with. Start free.

Available on phone, desktop & Chrome