Granted Patent B1

US 12,443,621 · App. 18/668,888 · Granted Oct 14, 2025

Inventors: Vibhor Porwal (Bidhuna, IN); Subrata Mitra (Bangalore, IN); Shubham Agarwal (West Bengal, IN); Ryan A Rossi (Santa Clara, CA); Ghazi Shazan Ahmad (West Bengal, IN); Manav Ketan Doshi (Maharashtra, IN); Syam Manoj Kumar Paila (Andhra Pradesh, IN)

Assignee: ADOBE INC.

G06F16/26

Abstract

Claims (54)

1. One or more non-transitory computer-readable media having a plurality of executable instructions embodied thereon, which, when executed by one or more processors, cause the one or more processors to perform a method comprising:

determining, based on applying a sample of a dataset to a data visualization recommendation model, statistical features of the sample for generating data visualizations of the sample;

determining, based on applying the statistical features and the dataset to a regression model, corresponding computational costs of the statistical features when applying the statistical features to the dataset;

determining, based on applying a computational budget, the sample of the dataset and the corresponding computational costs to a reinforcement learning model, recommended statistical features by causing the reinforcement learning model to:

select corresponding statistical features from the statistical features of the sample to add to the recommended statistical features until the computational budget is met by determining, based on a reward function, whether to explore a statistical feature from the statistical features of the sample or exploit a previously explored statistical feature from the statistical features of the sample; and

minimize a loss function between the recommended statistical features and the statistical features of the sample until the loss function is within a threshold value; and

causing display, based on applying the dataset and the recommended statistical features to the data visualization recommendation model, of a recommended data visualization.

2. The one or more non-transitory computer-readable media of claim 1 , the method further comprising:

training the regression model to compute the corresponding computational costs of the statistical features for different dataset sizes, wherein the corresponding computational costs of the statistical features of the sample correspond to a size of the dataset.

3. The one or more non-transitory computer-readable media of claim 1 , the method further comprising:

sequentially applying different samples of the dataset to the reinforcement learning model to minimize the loss function based on a combined score of the recommended statistical features.

4. The one or more non-transitory computer-readable media of claim 1 , the method further comprising:

minimizing the loss function between a combined score of the recommended statistical features and a corresponding combined score of the statistical features of the sample within the threshold value.

5. The one or more non-transitory computer-readable media of claim 1 , the method further comprising:

determining the reward function based on a change in a score of the recommended statistical features when adding a corresponding statistical feature to the recommended statistical features with respect to a corresponding computational cost of the statistical feature.

6. The one or more non-transitory computer-readable media of claim 1 , wherein determining the statistical features of the sample further comprises applying statistical functions of the data visualization recommendation model to the sample of the dataset.

7. The one or more non-transitory computer-readable media of claim 1 , wherein the computational budget is an amount of time and the corresponding computational costs of the statistical features correspond to an estimated amount of time to compute a corresponding statistical feature of the statistical features based on a size of the dataset.

8. A computer-implemented method comprising:

determining, based on applying samples of a dataset to a data visualization recommendation model, statistical features of each sample of the samples for generating data visualizations of each sample;

determining, based on applying the statistical features of each sample to a regression model, corresponding computational costs of the statistical features of each sample when applying the statistical features of each sample to the dataset;

determining, based on sequentially applying each sample to a reinforcement learning model with a computational budget and with the corresponding computational costs of the statistical features of each sample, recommended statistical features by causing the reinforcement learning model to:

for each sample, select corresponding statistical features from the statistical features of the sample to add to corresponding recommended statistical features for the sample until the computational budget is met by determining, based on a reward function, whether to explore a statistical feature from the statistical features of each of the samples or exploit a previously explored statistical feature from the statistical features of each of the samples;

for each sample, minimize a loss function between the corresponding recommended statistical features for the sample and the statistical features of the sample; and

sequentially apply each sample to the reinforcement learning model until the loss function is within a threshold value to determine the recommended statistical features; and

causing display of a data visualization based on applying the dataset and the recommended statistical features to the data visualization recommendation model.

9. The computer-implemented method of claim 8 , further comprising:

training the regression model to compute the corresponding computational costs of the statistical features for different dataset sizes, wherein the corresponding computational costs of the statistical features of each sample correspond to a size of the dataset.

10. The computer-implemented method of claim 8 , further comprising:

minimizing the loss function based on a combined score of the corresponding recommended statistical features.

11. The computer-implemented method of claim 8 , further comprising:

minimizing the loss function between a corresponding combined score of the corresponding recommended statistical features and a combined score of the statistical features within the threshold value.

12. The computer-implemented method of claim 8 , further comprising:

determining the reward function based on a change in a score of the recommended statistical features when adding a corresponding statistical feature to the corresponding recommended statistical features for the sample with respect to a corresponding computational cost of the statistical feature.

13. The computer-implemented method of claim 8 , wherein each of the samples is randomly sampled and each sample size of each of the samples increases in size when sequentially applying each sample to the reinforcement learning model.

14. The computer-implemented method of claim 8 , wherein determining the statistical features of each sample further comprises applying statistical functions of the data visualization recommendation model to each sample.

15. A computing system comprising:

a processor; and

a non-transitory computer-readable medium having stored thereon instructions that when executed by the processor, cause the processor to perform operations including:

determining, based on applying samples of a dataset to a data visualization recommendation model, statistical features of each sample of the samples for generating data visualizations of each sample;

causing a regression model to determine, based on the statistical features of each sample, corresponding computational costs of the statistical features of each sample when applying the statistical features of each sample to the dataset;

causing a reinforcement learning model to determine, based on sequentially applying each sample to the reinforcement learning model with a computational budget and with the corresponding computational costs of the statistical features of each sample, recommended statistical features by causing the reinforcement learning model to:

for each sample, select corresponding statistical features from the statistical features of the sample to add to corresponding recommended statistical features for the sample until the computational budget is met by determining, based on a reward function, whether to explore a statistical feature from the statistical features of each of the samples or exploit a previously explored statistical feature from the statistical features of each of the samples; and

for each sample, minimize a loss function between the corresponding recommended statistical features for the sample and the statistical features of the sample; and

sequentially apply each sample to the reinforcement learning model until the loss function is within a threshold value to determine the recommended statistical features; and

causing display, based on applying the dataset and the recommended statistical features to the data visualization recommendation model, of a data visualization.

16. The computing system of claim 15 , the operations further including:

training the regression model to compute the corresponding computational costs of the statistical features for different dataset sizes, wherein the corresponding computational costs of the statistical features of each sample correspond to a size of the dataset.

17. The computing system of claim 15 , the operations further including:

minimizing the loss function based on a combined score of the corresponding recommended statistical features.

18. The computing system of claim 15 , the operations further including:

minimizing the loss function between a corresponding combined score of the corresponding recommended statistical features and a combined score of the statistical features within the threshold value.

19. The computing system of claim 15 , the operations further including:

determining the reward function based on a change in a score of the recommended statistical features when adding a corresponding statistical feature to the corresponding recommended statistical features for the sample with respect to a corresponding computational cost of the statistical feature.

20. The computing system of claim 15 , wherein each of the samples is randomly sampled and each sample size of each of the samples increases in size when sequentially applying each sample to the reinforcement learning model.

Assignments (1)

References Cited (136)