Systems and methods for multimodal fusion of missing and unpaired image and tabular data for defect classification
A method may include providing a data set including rows of data. The rows of data may include at least one row of unpaired modality including a first modality, and at least one row of paired modality may include both the first modality and a second modality. The method may further include imputing, by a modality-specific encoder, the at least one row of unpaired modality by interpolating embeddings from the second modality of the paired modality; training, in a latent space, the modality-specific encoder based on the imputation for unimodal prediction and bimodal prediction; and generating a confidence value for the unimodal prediction and the bimodal prediction.
1 . A method, comprising:
providing a first data set comprising rows of data, the rows of data comprising at least one row of unpaired modality comprising a first modality, and at least one row of paired modality comprising both the first modality and a second modality;
imputing, by a modality-specific encoder, the at least one row of unpaired modality by interpolating embeddings from the second modality of the paired modality based on one or more prior embeddings of the second modality of the paired modality;
based on the imputing of the at least one row of the unpaired modality and based on a second set of data, training, in a latent space, the modality-specific encoder for a unimodal prediction of a missing modality and a bimodal prediction of the missing modality; and
generating a confidence value for the unimodal prediction and the bimodal prediction.
2 . The method of claim 1 , wherein the generating the confidence value comprises computing Shapley-based explanations for the unimodal prediction and the bimodal prediction.
3 . The method of claim 2 , wherein the computing the Shapley-based explanations comprises comparing an impact of the unimodal prediction and an impact of the bimodal prediction with a predetermined threshold.
4 . The method of claim 3 , further comprising selecting either the unimodal prediction or the bimodal prediction based on the generated confidence value.
5 . The method of claim 1 , wherein the second modality is a missing modality from the unpaired modality,
wherein the one or more prior embeddings of the second modality of the paired modality comprise K priors of the second modality, and
wherein the interpolating the embeddings comprises selecting the K priors of the second modality, the K priors being second modality embeddings of K samples of the at least one row of paired modality having closest embeddings of observed modality.
6 . The method of claim 5 , further comprising computing a weighted sum of the K priors by taking a cross-attention between the K samples and the K priors.
7 . The method of claim 1 , wherein the first modality corresponds to an image modality and the second modality corresponds to a tabular modality.
8 . The method of claim 1 , wherein the first modality correspond to a tabular modality and the second modality corresponds to an image modality.
9 . The method of claim 1 , further comprising training the modality-specific encoder for image modality by performing Vision Transformer.
10 . The method of claim 1 , further comprising training the modality-specific encoder for tabular data modality by performing Feature-Tokenizer Transformer.
11 . A system, comprising:
a memory; and
a processor configured to execute instructions stored in the memory to perform operations comprising:
providing a first data set comprising rows of data, the rows of data comprising at least one row of unpaired modality comprising a first modality, and at least one row of paired modality comprising both the first modality and a second modality;
imputing, by a modality-specific encoder, the at least one row of unpaired modality by interpolating embeddings from the second modality of the paired modality based on one or more prior embeddings of the second modality of the paired modality;
based on the imputing of the at least one row of the unpaired modality and based on a second set of data, training, in a latent space, the modality-specific encoder for a unimodal prediction of a missing modality and a bimodal prediction of the missing modality; and
generating a confidence value for the unimodal prediction and the bimodal prediction.
12 . The system of claim 11 , wherein the generating the confidence value comprises computing Shapley-based explanations for the unimodal prediction and the bimodal prediction.
13 . The system of claim 12 , wherein the computing the Shapley-based explanations comprises comparing an impact of the unimodal prediction and an impact of the bimodal prediction with a predetermined threshold.
14 . The system of claim 13 , wherein the operations further comprise selecting either the unimodal prediction or the bimodal prediction based on the generated confidence value.
15 . The system of claim 11 , wherein the second modality is a missing modality from the unpaired modality,
wherein the one or more prior embeddings of the second modality of the paired modality comprise K priors of the second modality, and
wherein the interpolating the embeddings comprises selecting the K priors of the second modality, the K priors being second modality embeddings of K samples of the at least one row of paired modality having closest embeddings of observed modality.
16 . The system of claim 15 , wherein the operations further comprise computing a weighted sum of the K priors by taking a cross-attention between the K samples and the K priors.
17 . The system of claim 11 , wherein the first modality corresponds to an image modality and the second modality corresponds to a tabular modality.
18 . The system of claim 11 , wherein the first modality correspond to a tabular modality and the second modality corresponds to an image modality.
19 . The system of claim 11 , wherein the operations further comprise training the modality-specific encoder for image modality by performing Vision Transformer.
20 . The system of claim 11 , wherein the operations further comprise training the modality-specific encoder for tabular data modality by performing Feature-Tokenizer Transformer.