CCSLM via the ReLU entry point - Flyer
CCSLM via the ReLU entry point
Break a ReLU neural network into a main data channel and a context channel.
ReLU(Wx) = DWx, where D is a diagonal decision matrix containing 0/1 entries produced by a context function applied to the layer input—in this case, the ReLU decisions.
The main data channel therefore only ever sees linear mappings: the W matrices and the D selection matrices. This is a remarkable restriction on apparent expressiveness, but it has potentially very positive consequences.
The context channel is doing something quite different. It does not transform the data; it selects which linear mapping acts on the data. A sequence of binary decisions across layers can represent an exponentially large number of distinct symbolic contexts or computational paths, even though each individual decision is simple.
The selected mapping can be made explicit by fusing D with an adjacent weight matrix. L = WD is the more useful representation for human comprehension: the context has selected columns of W, producing the linear mapping actually used by the data.
Also, D = DD: the selection operation is idempotent.
From this viewpoint, a ReLU neural network can be described as a:
Concatenated Context-Selected Linear Mapping (CCSLM)
This reframing separates two potentially independent engineering problems:
Context engineering — how computational regions are selected.
Linear mapping engineering — what each selected region learns to do.
That separation suggests many design opportunities beyond ReLU: alternative context sources, larger discrete context alphabets, external sensors, temporal state, hashing, automata, learned routing, or combinations of these. The context mechanism can potentially provide enormous symbolic complexity while leaving the data transformation itself strictly linear.

Comments
Post a Comment