1 unstable release

0.1.0 Apr 28, 2023

#425 in Machine learning

Custom license

17KB
447 lines

ThRust

ThRust is a Rust crate that provides a framework for thermodynamic and probabilistic computing. This package currently supports the following features:

  • Basic pbit gates: COPY, NOT, AND, OR.
  • Basic p-circuit compilation and composability.
  • p-circuit simulation through Markov Chain Monte Carlo sampling.

Tutorial

GridPbit's are used to construct a 2D lattice.

let p0 = GridPbit::new(0, 0);
let p1 = GridPbit::new(0, 1);

This construction can be visualized in the following way:

Pbits can then be used as inputs to probabilistic spin logic (PSL) gates:

let cp0 = Copy::new(p0, p1);

A set of PSL gates can be used to construct a p-circuit:

let circuit = Circuit::new();
circuit.append(Box::new(cp0));
circuit.compile();

Alternatively, circuits can be created from a vector of PSL gates:

let circuit = Circuit::from_vector(vec![Box::new(cp0)]);
circuit.compile();

Finally, create a new simulator to find the ground state configurations via Markov chain Monte Carlo

let mut sim = MCMC::new(circuit);
sim.run();

Composability

ThRust also implements basic p-circuit composability. For example, here's an example of a two-gate p-circuit:

let p1 = GridPbit::new(0, 0);
let p2 = GridPbit::new(0, 1);
let p3 = GridPbit::new(1, 0);
let p4 = GridPbit::new(2, 0);
let p5 = GridPbit::new(2, 1);

let mut circuit = Circuit::new();
circuit.append(Box::new(And::new(p1, p2, p3)));
circuit.append(Box::new(Or::new(p3, p4, p5)));
circuit.compile();

let mut sim = MCMC::new(circuit);
sim.run();

Future Additions

Running list of features to implement:

  • Circuit compilation for disjoint circuits
  • More robust algorithm for circuit compilation
  • Better plotting implementation with more options
  • Compute and report observables including confiuration energy
  • Enforce connectivity constraints
  • Compile high level code for FPGA hardware implementations

References

Various references to learn more about probabilistic computing.

  1. Thermodynamic AI and the fluctuation frontier
  2. p-Bits for Probabilistic Spin Logic
  3. Massively Parallel Probabilistic Computing with Sparse Ising Machines

Citation

If you use this repository, please cite!

License

See the LICENSE file.

Dependencies

~17MB
~162K SLoC