moose icon indicating copy to clipboard operation
moose copied to clipboard

Secure distributed dataflow framework for encrypted machine learning and data processing

Results 100 moose issues
Sort by recently updated
recently updated
newest added

Related: https://github.com/tf-encrypted/runtime/issues/462

Per https://github.com/tf-encrypted/runtime/pull/944#pullrequestreview-909196632 we might get a performance improvement by switching to another type of petgraph. This issue is about trying out that idea to see if it makes a difference.

We currently have large and similar `match op.kind` expressions in implementations of eg `Session::execute` and `SymbolicStrategy::execute`, all of which could be replaced by various `impl DispatchKernel for Operator` potentially generated...

`SymbolicSession::add_operation` is (perhaps unsurprisingly) responsible for a lot of memory allocations. This issue is about investigate what, if anything, can be done to optimize this.

It is not entirely clear that we actually want to do this, so a first step of this issue is to investigate whether it's worth it

- Identity op @ replicated (from host input) - logical or @ replicated (from host input) comp to reproduce: ``` @edsl.computation def my_comp(): with bob: x = edsl.constant(np.array([0, 1], dtype=np.bool))...

> `net2` crate has been deprecated; use `socket2` instead | Details | | | ------------------- | ---------------------------------------------- | | Status | unmaintained | | Package | `net2` | | Version...

The `Abstract` prefix (as in `AbstractReplicatedTensor` and `AbstractRingTensor`) was introduced because of `ReplicatedSetup`: we could use `ReplicatedTensor` instead of `AbstractReplicatedTensor` since it doesn't clash with `Replicated64Tensor` and `Replicated128Tensor`, but we...

At the moment, if a computations uses Save to save something and then Load to load it right back, there is a chance that the Save and Load ops would...