Interactive qualitative-quantitative live labeling for deep learning artificial intelligence
A live model of a deep learning algorithm may be used to generate predictions of features of interest in an instance of training data. If the predictions correspond to actual features of interest, the predictions may be converted to qualitative labels, the instance may be designated as being acceptably labeled, and the live model may be trained on all instances of training data designated as acceptably labeled to update the live model. If the predictions do not correspond, a repetitive process of applying qualitative labels to features of interest in the instance of training data, quantitatively training on the qualitatively labeled instance of training data and all instances of training data designated as acceptably labeled, and generating predictions of features of interest until the predictions correspond to the actual features of interest.
1. A method of interactive qualitative-quantitative live labeling a training dataset with real-time feedback while labeling each instance of training data to reduce an amount of time required to produce a complete labeled training dataset for training an intended model of a second deep learning algorithm for operative use, the method comprising:
for each instance of training data in the training dataset to be labeled:
submitting the instance of training data to a live model comprising a first deep learning algorithm in a live training mode;
generating one or more initial predictions of features of interest in an instance of training data in the training dataset with the live model;
converting the one or more initial predictions of features of interest into one or more provisional qualitative labels applied to the instance of training data;
receiving qualitative labels for one or more features of interest in the instance of training data from an operator, wherein the qualitative labels correspond to additions to, deletions from, or modifications of, the one or more provisional qualitative labels or acceptance of the one or more provisional qualitative labels, resulting in a qualitatively labeled instance of training data;
quantitatively training the live model on the qualitatively labeled instance of training data and all instances of training data previously designated as acceptably labeled to update the live model;
generating one or more predictions of features of interest in the instance of training data with the updated live model;
receiving, from the operator, a determination as to whether the one or more predictions of features of interest in the instance of training data are a same as the one or more qualitative labels applied to the instance of training data;
designating the instance of training data as being acceptably labeled if the one or more predictions are determined to be the same as the one or more qualitative labels, then qualitatively training the live model on the instance of training data and all other instances of training data designated as acceptably labeled to update the live model, and advancing to the next instance of training data; and
repeating the receiving qualitative labels, the qualitatively training the live model, the generating one or more predictions of features of interest, and the receiving from the operator the determination, until the determination is that the one or more predictions of features of interest are the same as the one or more qualitative labels applied to the instance of training data,
wherein the first deep learning algorithm of the live model has a depth, width, size, capacity, and complexity less than the second deep learning algorithm of the intended model, and
wherein an amount of time required to qualitatively label each instance of training data in the training dataset decreases over time as additional instances of training data are designated as acceptably labeled and used to update the live model.
2. The method of claim 1 , wherein the first deep learning algorithm of the live model comprises model parameters that have values that are initialized to zero, predetermined values, or random values.
3. The method of claim 1 , wherein the first deep learning algorithm of the live model comprises an artificial neural network, a radial basis function network, a convolutional neural network, a recurrent neural network, a long short-term memory network, a self-organizing map, an autoencoder, or a deep belief network.
4. The method of claim 1 , wherein the labeled training dataset comprises all instances of training data designated as acceptably labeled.
5. The method of claim 1 , wherein the second deep learning algorithm of the intended model comprises an artificial neural network, a radial basis function network, a convolutional neural network, a recurrent neural network, a long short-term memory network, a self-organizing map, an autoencoder, or a deep belief network.
6. The method of claim 1 , wherein the training dataset comprises tens of thousands, hundreds of thousands, or millions of instances of training data.
7. The method of claim 1 , wherein the training dataset comprises numeric, alphanumeric, or graphical data.
8. The method of claim 1 , wherein the training dataset comprises nominal, ordinal, discrete, continuous, or graphical data, or combinations thereof.