Identifying prompts used for training of inference models
Systems, methods and non-transitory computer readable media for identifying prompts used for training of inference models are provided. In some examples, a specific textual prompt in a natural language may be received. Further, data based on at least one parameter of an inference model may be accessed. The inference model may be a result of training a machine learning model using a plurality of training examples. Each training example of the plurality of training examples may include a respective textual content and a respective media content. The data and the specific textual prompt may be analyzed to determine a likelihood that the specific textual prompt is included in at least one training example of the plurality of training examples. A digital signal indicative of the likelihood that the specific textual prompt is included in at least one training example of the plurality of training examples may be generated.
1. A non-transitory computer readable medium storing a software program comprising data and computer implementable instructions that when executed by at least one processor cause the at least one processor to perform operations for identifying prompts used for training of inference models, the operations comprising:
receiving a specific textual prompt in a natural language;
accessing data based on at least one parameter of an inference model, the inference model is a result of training a machine learning model using a plurality of training examples, each training example of the plurality of training examples includes a respective textual content and a respective media content;
analyzing the data and the specific textual prompt to determine a likelihood that the specific textual prompt is included in at least one training example of the plurality of training examples; and
generating a digital signal indicative of the likelihood that the specific textual prompt is included in at least one training example of the plurality of training examples.
2. The non-transitory computer readable medium of claim 1 , wherein the likelihood is a binary likelihood, and wherein the determining the likelihood includes determining whether the specific textual prompt is included in at least one training example of the plurality of training examples.
3. The non-transitory computer readable medium of claim 1 , wherein the operations further comprise:
using the specific textual prompt and the at least one parameter of the inference model to obtain an output of the inference model corresponding to the specific textual prompt; and
analyzing the output to determine the likelihood.
4. The non-transitory computer readable medium of claim 1 , wherein the operations further comprise:
using the specific textual prompt to generate a plurality of variations of the specific textual prompt;
for each variation of the plurality of variations, using the variation and the at least one parameter of the inference model to obtain an output of the inference model corresponding to the variation; and
analyzing the outputs to determine the likelihood.
5. The non-transitory computer readable medium of claim 4 , wherein the operations further comprise:
obtaining a plurality of directions in a mathematical space;
obtaining a specific mathematical object in the mathematical space corresponding to the specific textual prompt;
for each direction of the plurality of directions, using the specific mathematical object and the direction to determine a mathematical object in the mathematical space corresponding to the specific mathematical object and the direction; and
for each direction of the plurality of directions, generating a textual prompt corresponding to the mathematical object corresponding to the specific mathematical object and the direction, thereby generating the plurality of variations of the specific textual prompt.
6. The non-transitory computer readable medium of claim 4 , wherein the operations further comprise:
selecting a plurality of utterances, no utterance of the plurality of utterances is included in the specific textual prompt; and
for each utterance in the plurality of utterances, analyzing the specific textual prompt to generate a variation of the specific textual prompt that includes the utterance, thereby generating the plurality of variations of the specific textual prompt.
7. The non-transitory computer readable medium of claim 4 , wherein the operations further comprise:
analyzing the specific textual prompt to detect a plurality of utterances included in the specific textual prompt; and
for each utterance in the plurality of utterances, analyzing the specific textual prompt to generate a variation of the specific textual prompt that do not include the utterance, thereby generating the plurality of variations of the specific textual prompt.
8. The non-transitory computer readable medium of claim 1 , wherein the operations further comprise:
using the specific textual prompt and the at least one parameter of the inference model to obtain a gradient, corresponding to the specific textual prompt, of a function associated with the inference model; and
analyzing the gradient to determine the likelihood.
9. The non-transitory computer readable medium of claim 1 , wherein the operations further comprise:
using the specific textual prompt and the at least one parameter of the inference model to calculate a loss corresponding to the specific textual prompt and to a loss function associated with the machine learning model; and
analyzing the loss to determine the likelihood.
10. The non-transitory computer readable medium of claim 1 , wherein the operations further comprise:
analyzing the data and the specific textual prompt to determine a second likelihood that a variation version of the specific textual prompt is included in at least one training example of the plurality of training examples; and
generating a second digital signal indicative of the second likelihood.
11. The non-transitory computer readable medium of claim 1 , wherein the inference model is a generative model, and wherein for each training example of the plurality of training examples, the respective textual content is a respective input textual prompt and the respective media content is a respective desired output media content.
12. The non-transitory computer readable medium of claim 1 , wherein the operations further comprise:
analyzing the data and the specific textual prompt to determine a measure of similarity of the specific textual prompt to the textual content included in a selected training example of the plurality of training examples; and
generating a second digital signal indicative of the measure of similarity of the specific textual prompt to the selected training example of the plurality of training examples.
13. The non-transitory computer readable medium of claim 12 , wherein the measure of similarity is indicative of an amount of variation.
14. The non-transitory computer readable medium of claim 12 , wherein the second digital signal is indicative of the selected training example of the plurality of training examples.
15. The non-transitory computer readable medium of claim 12 , wherein the textual content included in the selected training example is the most similar to specific textual prompt of all the textual contents included in the plurality of training examples.
16. The non-transitory computer readable medium of claim 1 , wherein the specific textual prompt includes a first noun, wherein the textual content included in a particular training example includes a second noun, and wherein the likelihood is based on the first noun and the second noun.
17. The non-transitory computer readable medium of claim 1 , wherein the specific textual prompt includes a particular noun and a first adjective adjacent to the particular noun, wherein the textual content included in a particular training example includes the particular noun and a second adjective adjacent to the particular noun, and wherein the likelihood is based on the first adjective and the second adjective.
18. The non-transitory computer readable medium of claim 1 , wherein the specific textual prompt includes a particular verb and a first adverb adjacent to the particular verb, wherein the textual content included in a particular training example includes the particular verb and a second adverb adjacent to the particular verb, and wherein the likelihood is based on the first adverb and the second adverb.
19. A method for identifying prompts used for training of inference models, the method comprising:
receiving a specific textual prompt in a natural language;
accessing data based on at least one parameter of an inference model, the inference model is a result of training a machine learning model using a plurality of training examples, each training example of the plurality of training examples includes a respective textual content and a respective media content;
analyzing the data and the specific textual prompt to determine a likelihood that the specific textual prompt is included in at least one training example of the plurality of training examples; and
generating a digital signal indicative of the likelihood that the specific textual prompt is included in at least one training example of the plurality of training examples.
20. A system for identifying prompts used for training of inference models, the system comprising:
at least one processor configured to perform the operations of:
receiving a specific textual prompt in a natural language;
accessing data based on at least one parameter of an inference model, the inference model is a result of training a machine learning model using a plurality of training examples, each training example of the plurality of training examples includes a respective textual content and a respective media content;
analyzing the data and the specific textual prompt to determine a likelihood that the specific textual prompt is included in at least one training example of the plurality of training examples; and
generating a digital signal indicative of the likelihood that the specific textual prompt is included in at least one training example of the plurality of training examples.