Systems and methods for adaptive preprocessor selection for efficient multi-modal classification

A method for minimizing computational cost of preprocessors when analyzing multi-modal data in real-time, the method comprising:

• receiving a multi-modal input using one or more sensors, wherein the multi-modal input comprises a multi-modal data format;
• accessing a cache storing a plurality of previously analyzed inputs and a corresponding preprocessor for each previously analyzed input;
• determining, based on a similarity metric, a closest previously analyzed input having first values closest to second values of the multi-modal input;
• selecting a first preprocessor based on the corresponding preprocessor of the closest previously analyzed input;
• inputting the multi-modal input into the first preprocessor to generate a first output, wherein the first output comprises a single modal data format, and wherein the first preprocessor has a first computational cost;
• inputting the first output into an artificial intelligence model, wherein the artificial intelligence model is trained to determine classifications of inputted data and determine confidence measures for each respective classification;
• receiving a first model output from the artificial intelligence model, wherein the first model output indicates a first classification for the first model output and a first confidence measure for the first classification;
• comparing the first confidence measure to a threshold confidence measure;
• determining that the first confidence measure does not correspond to the threshold confidence measure;
• in response to determining that the first confidence measure does not correspond to the threshold confidence measure, selecting a second preprocessor from a plurality of preprocessors based on the second preprocessor having a greater computational cost than the first preprocessor;
• inputting the multi-modal input into the second preprocessor to generate a second output, wherein the second output comprises the single modal data format;
• and
• generating for display, on a user interface, a recommendation based on the second output.

A One or more non-transitory computer-readable media for detecting sensitive information from multi-modal inputs, comprising instructions that, when executed by one or more processors, cause the one or more processors to perform operations comprising:

• receiving a multi-modal input using one or more sensors, wherein the multi-modal input comprises a multi-modal data format;
• accessing a cache storing a plurality of previously analyzed inputs and a corresponding preprocessor for each previously analyzed input;
• determining, based on a similarity metric, a closest previously analyzed input having first values closest to second values of the multi-modal input;
• selecting a first preprocessor based on the corresponding preprocessor of the closest previously analyzed input;
• inputting the multi-modal input into the first preprocessor to generate a first output, wherein the first output comprises a single modal data format, and wherein the first preprocessor has a first computational cost;
• inputting the first output into an artificial intelligence model, wherein the artificial intelligence model is trained to determine classifications of inputted data and determine confidence measures for each respective classification to obtain a first model output, wherein the first model output indicates a first classification for the first model output and a first confidence measure for the first classification;
• in response to determining that the first confidence measure does not correspond to a threshold confidence measure, selecting a second preprocessor having a greater computational cost than the first preprocessor from a plurality of preprocessors;
• inputting the multi-modal input into the second preprocessor to generate a second output, wherein the second output comprises the single modal data format;
• and
• generating for display, on a user interface, a recommendation based on the second output.