Data candidate querying via embeddings for deep learning refinement
Systems/techniques that facilitate data candidate querying via embeddings for deep learning refinement are provided. In various embodiments, a system can access a test data candidate provided by a client, generate, via a first deep learning neural network, an inferencing output based on the test data candidate, and access feedback indicating whether the client accepts or rejects the inferencing output. In various aspects, the system can generate, via at least one second deep learning neural network, at least one embedding based on the test data candidate. In various instances, the system can, in response to the feedback indicating that the client rejects the inferencing output, identify, in a candidate-embedding dataset, one or more data candidates whose embeddings are within a threshold level of similarity to the at least one embedding and can retrain the first deep learning neural network based on the one or more data candidates.
1 . A system, comprising:
a memory configured to store computer-executable components; and
a processor that executes the computer-executable components that:
generates, using a first deep learning neural network performing an inferencing task, an inferencing output based on a client data candidate provided by a client, wherein the client data candidate comprises medical image data of a medical patient, wherein the first deep learning neural network has previously been trained to perform the inferencing task using a set of training data candidates, wherein the client data candidate has a privacy restriction that limits access by the first deep learning neural network and a second deep learning neural network to the client data candidate for a defined time period to perform the inferencing task;
generates, via the second deep learning neural network during the defined time period, a first embedding based on the client data candidate, wherein the first embedding is a vectorized representation of the client data candidate that is not decipherable by the system, the first deep learning neural network, nor the second deep learning neural network to obtain the client data candidate;
receives, from the client, after the defined time period, feedback indicating that the inferencing output is faulty;
identifies, from a candidate-embedding dataset comprising data candidates having respective second embeddings generated by the second deep learning neural network, one or more data candidates having respective second embeddings that are within a threshold level of similarity to the first embedding based on at least one metric, wherein in the data candidates have respective ground truth annotations associated with the inferencing task; and
retrains, using the one or more data candidates, the first deep learning neural network to perform the inferencing task.
2 . The system of claim 1 , wherein the candidate-embedding dataset is stored in a cloud computing environment.
3 . The system of claim 1 , wherein the candidate-embedding dataset is stored in a non-cloud computing environment.
4 . The system of claim 1 , wherein the inferencing output is a classification, segmentation, or regression based on the medical image data, and wherein the vectorized representation has a smaller dimensionality than the medical image data.
5 . The system of claim 1 , wherein the at least one metric comprises at least one of a cosine similarity or a Euclidean distance.
6 . The system of claim 1 , wherein the at least one of the computer-executable components further:
trains the second deep learning neural network in an unsupervised fashion to generate embeddings of the data candidates.
7 . The system of claim 1 , wherein the at least one of the computer-executable components further:
trains the second deep learning neural network in a supervised fashion to generate embeddings of the data candidates.
8 . A computer-implemented method, comprising:
generating, by a system comprising a processor, using a first deep learning neural network performing an inferencing task, an inferencing output based on a client data candidate provided by a client, wherein the client data candidate comprises medical image data of a medical patient, wherein the first deep learning neural network has previously been trained to perform the inferencing task using a set of training data candidates, wherein the client data candidate has a privacy restriction that limits access by the first deep learning neural network and a second deep learning neural network to the client data candidate for a defined time period to perform the inferencing task;
generating, by the system, using the second deep learning neural network during the defined time period, a first embedding based on the client data candidate, wherein the first embedding is a vectorized representation of the client data candidate that is not decipherable by the system, the first deep learning neural network, nor the second deep learning neural network to obtain the client data candidate;
receiving, by the system, from the client, after the defined time period, feedback indicating that the inferencing output is faulty;
identifying, by the system, from a candidate-embedding dataset comprising data candidates having respective second embeddings generated by the second deep learning neural network, one or more data candidates having respective second embeddings that are within a threshold level of similarity to the first embedding based on at least one metric, wherein in the data candidates have respective ground truth annotations associated with the inferencing task; and
retraining, by the system, using the one or more data candidates, the first deep learning neural network to perform the inferencing task.
9 . The computer-implemented method of claim 8 , wherein the candidate-embedding dataset is stored in a cloud computing environment.
10 . The computer-implemented method of claim 8 , wherein the candidate-embedding dataset is stored in a non-cloud computing environment.
11 . The computer-implemented method of claim 8 , wherein the inferencing output is a classification, segmentation, or regression based on the medical image data, and wherein the vectorized representation has a smaller dimensionality than the medical image data.
12 . The computer-implemented method of claim 8 , wherein the at least one metric comprises at least one of a cosine similarity or Euclidean distance.
13 . The computer-implemented method of claim 8 , further comprising:
training, by the system, the second deep learning neural network in an unsupervised fashion to generate embeddings of the data candidates.
14 . The computer-implemented method of claim 8 , further comprising:
training, by the system, the second deep learning neural network in a supervised fashion to generate embeddings of the data candidates.
15 . A computer program product comprising a non-transitory computer-readable memory having program instructions embodied therewith, the program instructions executable by a processor of a system to cause the processor to:
generate, using a first deep learning neural network performing an inferencing task, an inferencing output based on a client data candidate provided by a client, wherein the client data candidate comprises medical image data of a medical patient, wherein the first deep learning neural network has previously been trained to perform the inferencing task using a set of training data candidates, wherein the client data candidate has a privacy restriction that limits access by the first deep learning neural network and a second deep learning neural network to the client data candidate for a defined time period to perform the inferencing task;
generate, using the second deep learning neural network during the defined time period, a first embedding based on the client data candidate, wherein the first embedding is a vectorized representation of the client data candidate that is not decipherable by the system, the first deep learning neural network, nor the second deep learning neural network to obtain the client data candidate;
identify, from a candidate-embedding dataset comprising data candidates having respective second embeddings generated by the second deep learning neural network, one or more data candidates having respective second embeddings that are within a threshold level of similarity to the first embedding based on at least one metric, wherein in the data candidates have respective ground truth annotations associated with the inferencing task; and
retrain, using the one or more data candidates, the first deep learning neural network to perform the inferencing task.
16 . The computer program product of claim 15 , wherein the candidate-embedding dataset is stored in a cloud computing environment.
17 . The computer program product of claim 15 , wherein the inferencing output is a classification, segmentation, or regression based on the medical image data, and wherein the vectorized representation has a smaller dimensionality than the medical image data.
18 . The computer program product of claim 15 , wherein the at least one metric comprises at least one of a cosine similarity or Euclidean distance.
19 . The computer program product of claim 15 , wherein the program instructions are further executable to cause the processor to:
train the second deep learning neural network in an unsupervised fashion to generate embeddings of the data candidates.
20 . The computer program product of claim 15 , wherein the program instructions are further executable to cause the processor to:
train the second deep learning neural network in a supervised fashion to generate embeddings of the data candidates.