Techniques for improving machine-learning accuracy and convergence

1. A computer-implemented method, comprising:

obtaining, by a computing device, a machine-learning model that has been previously trained to identify an image vector from an input image;

obtaining, by the computing device, a set of image feature vectors based at least in part on providing individual images of a set of images to the machine-learning model as input, the set of images comprising varying images of an item;

calculating, by the computing device, a similarity score for each pair of image feature vectors from the set of image feature vectors obtained from the machine-learning model;

presenting, at a user interface, the similarity score for each pair of image feature vectors from the set of image feature vectors;

receiving, via the user interface, user input indicating removal of a specific image from the set of images;

in response to the user input, removing the image from the set of images;

updating the user interface based at least in part on removing the image from the set of images; and

conducting at least one user interface experiment utilizing the set of images as updated.

2. The computer-implemented method of claim 1 , wherein presenting the similarity score comprises presenting a grid of similarity scores that comprises the similarity score.

3. The computer-implemented method of claim 2 , further comprising highlighting one or more areas of the grid based at least in part on determining one or more corresponding similarity scores associated with the one or more areas exceeds a predefined threshold value.

4. The computer-implemented method of claim 1 , further comprising:

obtaining, by the computing device, an additional machine-learning model that has been previously trained to identify a textual feature vector from input text;

obtaining, by the computing device, a set of textual feature vectors based at least in part on providing individual instances of text of a set of textual examples to the machine-learning model as input;

calculating, by the computing device, an additional similarity score for each pair of instances of text from the set of textual feature vectors obtained from the machine-learning model;

presenting, at the user interface, the similarity score for each pair of textual feature vectors from the set of textual feature vectors; and

receiving additional user input indicating a subset of textual examples from the set of textual examples, wherein the at least one user interface experiment is conducted utilizing the subset of textual examples with the set of images as updated.

5. A computing device comprising:

one or more processors; and

one or more memories storing computer-executable instructions that, when executed with the one or more processors, cause the computing device to:

obtain a first set of item assets of a first dimension, the first set of item assets being associated with an item;

obtain a first set of vectors corresponding to the first set of item assets, the first set of vectors comprising feature embeddings of the first set of item assets;

determine a degree of similarity between a pair of vectors of the first set of vectors;

present, at a user interface a representation indicating the degree of similarity between the pair of vectors;

receive, at the user interface, user input reducing the first set of item assets to a subset of the first set of item assets; and

conduct at least one experiment utilizing the subset of the first set of item assets, the at least one experiment being conducted utilizing the subset of the first set of item assets based at least in part the user input.

6. The computing device of claim 5 , wherein the computing device is further caused to:

obtain a second set of item assets of a second dimension different from the first dimension, the second set of item assets being associated with the item;

obtain a second set of vectors corresponding to the second set of item assets, the second set of vectors comprising respective feature embeddings of the second set of item assets;

determine a second degree of similarity between a second pair of vectors of the second set of vectors;

present, at the user interface, a second representation indicating the second degree of similarity between the second pair of vectors at; and

receive, at the user interface, second user input reducing the second set of item assets to a second subset of the second set of item assets, wherein the at least one experiment is conducted further utilizing the second subset of the second set of item assets.

7. The computing device of claim 6 , wherein the first dimension is an image dimension and the second dimension is a textual dimension.

8. The computing device of claim 6 , wherein the second set of item assets comprises at least one of: a plurality of item title variants, a plurality of item description variants, or a plurality of images of the item.

9. The computing device of claim 5 , wherein the representation is a similarity score that is calculated based at least in part on calculating structural similarity index between the pair of vectors.

10. The computing device of claim 9 , wherein the similarity score is presented as one of a plurality of similarity scores, the plurality of similarity scores corresponding to varying pairs of the first set of item assets.

11. The computing device of claim 5 , wherein the computing device is further caused to present the representation of the degree of similarity differently with respect to other representations of similarity corresponding to other pairs of item assets, wherein the representation is presented differently based at least in part on determining that the degree of similarity between the pair of vectors of the first set of vectors meets a threshold condition.

12. The computing device of claim 5 , wherein the user input received at the user interface comprises one or more selections indicating a removal of a corresponding item asset from the first set of item assets.

13. The computing device of claim 12 , wherein the at least one experiment is performed utilizing a multi-armed bandit algorithm.

14. A non-transitory computer readable storage medium comprising computer-executable instructions that, when executed with one or more processors of a computing device, cause the computing device to perform operations comprising:

obtaining a first set of vectors corresponding to a first set of item assets of a first dimension corresponding to an item, the first set of vectors comprising feature embeddings corresponding to the first set of item assets;

obtaining a second set of vectors corresponding to a second set of item assets of a second dimension corresponding to the item, the second dimension being different from the first dimension, the second set of vectors comprising additional feature embeddings corresponding to the second set of item assets;

presenting, via a first user interface element, a first representation indicating a first degree of similarity between a first pair of vectors of the first set of vectors;

presenting, via a second user interface element, a second representation indicating a second degree of similarity between a second pair of vectors of the second set of vectors;

receiving user input at the first user interface element and the second user interface element, the user input indicating a reduction in the first set of item assets or the second set of item assets; and

conducting at least one experiment utilizing a first subset of item assets from the first set of item assets and a second subset of item assets from the second set of item assets, the first subset of item assets and the second subset of item assets being identified based at least in part on the user input.

15. The non-transitory computer readable storage medium of claim 14 , wherein conducting the at least one experiment comprises:

presenting the first subset of item assets to varying sets of users at a webpage;

monitoring for subsequent user input received at the webpage; and

selecting, from the first subset of item assets, a particular item asset based at least in part on monitoring the subsequent user input.

16. The non-transitory computer readable storage medium of claim 15 , wherein conducting the at least one experiment comprises:

presenting the second subset of item assets at the webpage to various users;

monitoring for additional user input received at the webpage and corresponding to the second subset of item assets; and

selecting, from the second subset of item assets, a specific item asset based at least in part on monitoring the additional user input.

17. The non-transitory computer readable storage medium of claim 16 , wherein the computing device performs further operations comprising, indicating at a subsequent user interface, the particular item asset of the first set of item assets and the specific item asset from the second set of item assets.

18. The non-transitory computer readable storage medium of claim 14 , wherein the first set of item assets comprises a set of images and wherein the second set of item assets comprise a set of textual segments.

19. The non-transitory computer readable storage medium of claim 14 , wherein the first representation is a first similarity score that is calculated based at least in part on calculating a Euclidean distance between the first pair of vectors, and wherein the second representation is a second similarity score that is calculated based at least in part on calculating an additional Euclidean distance between the second pair of vectors.

20. The non-transitory computer readable storage medium of claim 14 , wherein conducting the at least one experiment utilizing the first subset of item assets from the first set of item assets causes a single item asset of the first subset of item assets to be selected from the first subset of item assets and presented to a user.