Table 1.
Some selected works, in no particular order, showing the principal approach of domain knowledge inclusion into deep neural networks. For each work referred here, we show the type of learner with following acronyms: Multilayer Perceptron (MLP), Convolutional Neural Network (CNN), Recurrent Neural Network (RNN), Graph Neural Network (GNN), Adaptive Resonance Theory-based Network Map (ARTMAP), DNN refers to a DNN structure dependent on intended task. We use ‘MLP’ here to represent any neural network, that conforms to a layered-structure that may or maynot be fully-connected. RNN also refers to sequence models constructed using Long Short-Term Memory (LSTM) or Gated Recurrent Unit (GRU) cells.
| Principal approach | Work (reference) | Type of learner |
|---|---|---|
| Transforming Data | DRM24,25 | MLP |
| CILP++28 | MLP | |
| R-GCN46 | GNN | |
| KGCN61 | GNN | |
| KBRD49 | GNN | |
| DG-RNN44 | RNN | |
| DreamCoder32 | DNN | |
| Gated-K-BERT38 | Transformer | |
| VEGNN5 | GNN | |
| BotGNN6 | GNN | |
| KRISP45 | GNN, Transformer | |
| Transforming Loss | IPKFL78 | CNN |
| ILBKRME30 | MLP | |
| HDNNLR64 | CNN, RNN | |
| SBR68 | MLP | |
| SBR69 | CNN | |
| DL266 | CNN | |
| Semantic Loss63 | CNN | |
| LENSR91 | GNN | |
| DANN67 | MLP | |
| PC-LSTM72 | RNN | |
| DomiKnowS73 | DNN* | |
| MultiplexNet74 | MLP, CNN | |
| Analogy Model75 | RNN | |
| Transforming Model | KBANN86 | MLP |
| Cascade-ARTMAP89 | ARTMAP | |
| CILP29 | RNN | |
| DeepProbLog93 | CNN | |
| LRNN101 | MLP | |
| TensorLog103 | MLP | |
| Domain-Aware BERT82 | Transformer | |
| NeuralLog104 | MLP | |
| DeepStochLog94 | DNN* |