Unverified50% confidenceFactExact time
D²NN arranges multiple layers of trainable diffractive planes, with the transmission or reflection coefficient of each micro-element serving as a learnable parameter; after optimization, these are fixed as a physical structure, and the diffraction and interference that occur as light passes through equivalently perform forward inference.
1
Sources
50%
Confidence
Long-term
Relevance
10/6/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedThe Dyna architecture simultaneously performs three things—direct learning, model learning, and planning—unifying the use of real and simulated experience.66% similarUnverifiedNeural network models such as Deep Knowledge Tracing (DKT) do not explicitly model mastery as a latent state, but instead directly predict the probability of a student's correct response to the next question through architectures such as recurrent neural networks or Transformers.65% similarUnverifiedQuantization-aware training inserts fake quantization nodes during the forward pass to simulate quantization noise, and during backpropagation uses the straight-through estimator (STE) to approximate gradients because the rounding operation is non-differentiable.65% similarUnverifiedThe inference engine uses a DMA streaming load strategy that prefetches the weights of the next layer from Flash storage via DMA while computing the current layer, in order to hide slow Flash read latency.64% similarUnverifiedGNoME基于图神经网络（GNN），通过学习Materials Project等密度泛函理论（DFT）数据库中已知晶体结构稳定性规律进行预测63% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/979825API
curl https://kongchang.com/api/v1/knowledge/claims/979825MCP
get_claim(id=979825)