Release v0.2.0

v0.2.0

• Use random session id on calls to AsyncTestRuntime via python bindings.
• Switch to FuturesUnordered for more efficient wait on computation completion
• Support arbitrary fixedpoint precision pairs from python bindings
• More complete support of tensor types kernels
• AsyncSession is now Send (can be sent between threads)
• Enable use of Shape values on non-Host placement types
• Add ZerosOp on a host placement (create a vector of zeros of a given shape)
• Add Tensorflow Keras support
• Add ReLU activation function (for multilayer perceptron and NN)

Release v0.1.6

• Memory usage optimizations and reduced cloning
• Wellformed compiler pass to statically verify a computation with elk: elk compile -p wellformed comp.moose
• TLS support for gRCP networking
• PyTorch neural networks support
• Elk now supports all three computation formats (textual, msgpack, and bincode)
• Ensures tensors are in “standard layout” (i.e. contiguous & row-major)
• Added local file storage to moose modules (reading csv and numpy data files supported)
• start_server convenance method added to gRPC network plugin

0.1.5

• Significant RAM usage improvements throughout the codebase
• Add Multilayer Perceptron (MLP), a type of neural network to pymoose predictors
• PrimDeriveSeed operator renamed to DeriveSeed
• PrimPrfKeyGen operator renamed to PrfKeyGen
• Added Host prefix to Seed, PrfKey and Unit
• Unified MeanOp and fuse RepFixedpointMeanOp with RingFixedpointMeanOp
• Using BitVec as the data storage for HostBitTensor
• Replace sodium oxide with thread_rng and blake3
• Fix binary LinearClassifier logic
• Uniform textual format for rendezvous and sync keys

v0.1.4

Less verbose error handling

v0.1.2

This is the first "stable" release with a complete set of public APIs.
Future releases are expected to contain instructions for migration from the previous "stable" version.

Changed

• Symbolic compilation support
• Sufficient primitives to execute AES decryption
• Sufficient primitives to execute custom models, including linear regression and XGBoost decision trees.