AI 용어집
인공지능 완전 사전
Dynamic Memory Networks (DMN)
Neural network with an episodic memory module that is updated sequentially at each time step to track the state of an input sequence.
Gated Recurrent Units (GRU) with external memory
Extension of GRUs where the hidden state is complemented by interactions with external memory, improving the ability to model very long-term dependencies.
Neural Random Access Machine (NRAM)
Architecture where a neural controller learns to generate instructions to manipulate a random-access memory (RAM), simulating the operation of a computer.
Read-Write Memory Networks
Models that explicitly separate read and write operations on memory, allowing finer control and more advanced reasoning capabilities.
Composable Memory Networks
Architecture where memory operations are designed as composable modules, enabling the construction of complex and flexible reasoning pipelines.
Differentiable Stack
Implementation of a stack structure (LIFO) whose push and pop operations are differentiable, which can be integrated into a neural network and trained by backpropagation.
Neural Programmer-Interpreters (NPI)
Hierarchical model where a neural controller learns to generate high-level programs that are then executed by an interpreter to manipulate external memory.