Multi-level time series forecasting using artificial intelligence techniques
Methods, apparatus, and processor-readable storage media for multi-level time series forecasting using artificial intelligence techniques are provided herein. An example computer-implemented method includes determining entity-related features and temporal features from at least a portion of one or more sets of time series data pertaining to at least one entity; creating multiple embeddings by encoding at least a portion of the entity-related features and at least a portion of the temporal features using at least one artificial intelligence-based embedding technique; processing the multiple embeddings using at least one neural network-based attention technique; generating one or more data forecasts across one or more temporal granularity levels by processing at least a portion of results from the processing of the multiple embeddings using at least one artificial intelligence-based categorization technique; and performing one or more automated actions based at least in part on the one or more data forecasts.
1 . A computer-implemented method comprising:
determining one or more entity-related features and one or more temporal features from at least a portion of one or more sets of time series data pertaining to at least one entity;
creating multiple embeddings by encoding at least a portion of the one or more entity-related features and at least a portion of the one or more temporal features using at least one embedding-based neural network comprising at least one embedding layer;
processing the multiple embeddings using at least one multi-head attention-based neural network comprising at least one intra-sample attention layer, which generates at least one weighted scalar product from one or more portions of the multiple embeddings, and at least one additional attention layer, which generates, using the at least one weighted scalar product, one or more eigenvectors and one or more corresponding eigenvalues;
generating one or more data forecasts across one or more temporal granularity levels by processing at least a portion of results from the processing of the multiple embeddings using at least one categorization-based neural network comprising at least one feed forward layer; and
performing one or more automated actions based at least in part on the one or more data forecasts;
wherein the method is performed by at least one processing device comprising a processor coupled to a memory.
2 . The computer-implemented method of claim 1 , wherein processing at least a portion of results from the processing of the multiple embeddings using at least one categorization-based neural network comprises processing at least a portion of the results from the processing of the multiple embeddings using at least one one-shot learning technique.
3 . The computer-implemented method of claim 1 , wherein performing one or more automated actions comprises automatically training the at least one categorization-based neural network based at least in part on the one or more data forecasts.
4 . The computer-implemented method of claim 1 , wherein performing one or more automated actions comprises automatically initiating at least one action, in connection with one or more external systems, in response to at least a portion of the one or more data forecasts.
5 . The computer-implemented method of claim 1 , wherein determining one or more entity-related features and one or more temporal features comprises selecting fractions of time series data across multiple samples within the one or more sets of time series data.
6 . The computer-implemented method of claim 1 , further comprising:
training the at least one embedding-based neural network based at least in part on combining inter-sample information and intra-sample information across multiple samples within the one or more sets of time series data.
7 . The computer-implemented method of claim 1 , wherein generating the one or more data forecasts comprises processing the at least a portion of results from the processing of the multiple embeddings using at least one masked eigen-score-based attention mechanism in conjunction with one or more one-shot learning prediction components.
8 . A non-transitory processor-readable storage medium having stored therein program code of one or more software programs, wherein the program code when executed by at least one processing device causes the at least one processing device:
to determine one or more entity-related features and one or more temporal features from at least a portion of one or more sets of time series data pertaining to at least one entity;
to create multiple embeddings by encoding at least a portion of the one or more entity-related features and at least a portion of the one or more temporal features using at least one embedding-based neural network comprising at least one embedding layer;
to process the multiple embeddings using at least one multi-head attention-based neural network comprising at least one intra-sample attention layer, which generates at least one weighted scalar product from one or more portions of the multiple embeddings, and at least one additional attention layer, which generates, using the at least one weighted scalar product, one or more eigenvectors and one or more corresponding eigenvalues;
to generate one or more data forecasts across one or more temporal granularity levels by processing at least a portion of results from the processing of the multiple embeddings using at least one categorization-based neural network comprising at least one feed forward layer; and
to perform one or more automated actions based at least in part on the one or more data forecasts.
9 . The non-transitory processor-readable storage medium of claim 8 , wherein processing at least a portion of results from the processing of the multiple embeddings using at least one categorization-based neural network comprises processing at least a portion of the results from the processing of the multiple embeddings using at least one one-shot learning technique.
10 . The non-transitory processor-readable storage medium of claim 8 , wherein performing one or more automated actions comprises automatically training the at least one categorization-based neural network based at least in part on the one or more data forecasts.
11 . The non-transitory processor-readable storage medium of claim 8 , wherein performing one or more automated actions comprises automatically initiating at least one action, in connection with one or more external systems, in response to at least a portion of the one or more data forecasts.
12 . The non-transitory processor-readable storage medium of claim 8 , wherein determining one or more entity-related features and one or more temporal features comprises selecting fractions of time series data across multiple samples within the one or more sets of time series data.
13 . The non-transitory processor-readable storage medium of claim 8 , wherein the program code when executed by the at least one processing device further causes the at least one processing device:
to train the at least one embedding-based neural network based at least in part on combining inter-sample information and intra-sample information across multiple samples within the one or more sets of time series data.
14 . An apparatus comprising:
at least one processing device comprising a processor coupled to a memory;
the at least one processing device being configured:
to determine one or more entity-related features and one or more temporal features from at least a portion of one or more sets of time series data pertaining to at least one entity;
to create multiple embeddings by encoding at least a portion of the one or more entity-related features and at least a portion of the one or more temporal features using at least one embedding-based neural network comprising at least one embedding layer;
to process the multiple embeddings using at least one multi-head attention-based neural network comprising at least one intra-sample attention layer, which generates at least one weighted scalar product from one or more portions of the multiple embeddings, and at least one additional attention layer, which generates, using the at least one weighted scalar product, one or more eigenvectors and one or more corresponding eigenvalues;
to generate one or more data forecasts across one or more temporal granularity levels by processing at least a portion of results from the processing of the multiple embeddings using at least one categorization-based neural network comprising at least one feed forward layer; and
to perform one or more automated actions based at least in part on the one or more data forecasts.
15 . The apparatus of claim 14 , wherein processing at least a portion of results from the processing of the multiple embeddings using at least one categorization-based neural network comprises processing at least a portion of the results from the processing of the multiple embeddings using at least one one-shot learning technique.
16 . The apparatus of claim 14 , wherein performing one or more automated actions comprises automatically training the at least one categorization-based neural network based at least in part on the one or more data forecasts.
17 . The apparatus of claim 14 , wherein performing one or more automated actions comprises automatically initiating at least one action, in connection with one or more external systems, in response to at least a portion of the one or more data forecasts.
18 . The apparatus of claim 14 , wherein determining one or more entity-related features and one or more temporal features comprises selecting fractions of time series data across multiple samples within the one or more sets of time series data.
19 . The apparatus of claim 14 , wherein the at least one processing device is further configured:
to train the at least one embedding-based neural network based at least in part on combining inter-sample information and intra-sample information across multiple samples within the one or more sets of time series data.
20 . The apparatus of claim 14 , wherein generating the one or more data forecasts comprises processing the at least a portion of results from the processing of the multiple embeddings using at least one masked eigen-score-based attention mechanism in conjunction with one or more one-shot learning prediction components.