![]() |
Deepity
Ultra-fast, SIMD-accelerated Predictive Coding
|
Convolutional counterpart to SimplePCNetwork, now device-aware (matches DirectKPPCNetwork's own port earlier tonight: DeviceType constructor parameter, backend member, cpuArena/gpuArena split). More...
#include <vector>#include <memory>#include <random>#include <deepity/layers/SimpleConvPCLayer.h>#include <deepity/utils/MemoryArena.h>#include <deepity/utils/DeviceMemoryArena.h>#include <deepity/backend/IComputeBackend.h>#include <deepity/backend/DeviceType.h>Go to the source code of this file.
Classes | |
| class | Deep::SimpleConvPCNetwork |
| Convolutional Predictive Coding Network built from SimpleConvPCLayer – precision-free, AdamW-capable, device-aware. More... | |
Namespaces | |
| namespace | Deep |
| This file exists to provide both Tensor.h and IComputeBackend.h with device types without introducing circular dependencies. | |
Convolutional counterpart to SimplePCNetwork, now device-aware (matches DirectKPPCNetwork's own port earlier tonight: DeviceType constructor parameter, backend member, cpuArena/gpuArena split).
Two real bugs fixed during this port, same class found and fixed in SimplePCNetwork/DirectKPPCNetwork: ProjectForward() used std::memcpy directly (wrong once beliefs/mu are device pointers) and had no isClamped guard (would overwrite an already-clamped terminal layer's z with a stale forward-projected value); Predict()/ PredictWithProjection() used the std::vector iterator-range constructor directly on beliefs (dereferences immediately, wrong for device memory).