US 12,437,235 B2
Auto sampling in internet-of-things analytics system via cached recycle BINS753634
Gandhi Sivakumar, Bentleigh (AU); Kushagra Manglik, Lucknow (IN); Vijay Ekambaram, Chennai (IN); and Luke Peter Macura, Lucas (AU)
Assigned to International Business Machines Corporation, Armonk, NY (US)
Filed by International Business Machines Corporation, Armonk, NY (US)
Filed on Nov. 29, 2021, as Appl. No. 17/536,336.
Prior Publication US 2023/0169390 A1, Jun. 1, 2023
Int. Cl. G06N 20/00 (2019.01); G06N 20/20 (2019.01); G06N 3/08 (2023.01)
CPC G06N 20/00 (2019.01) [G06N 20/20 (2019.01); G06N 3/08 (2013.01)] 11 Claims
OG exemplary drawing
 
1. A computer-implemented method comprising:
storing, in a cached recycle bin, a set of sensor data that is sent from a sensor at a transmit frequency;
sampling at least a portion of the set of sensor data from the cached recycle bin based on a sampling frequency and training a machine learning model using the sampled data;
in response to detecting that a performance of the machine learning model falls below a threshold during the training, adjusting the sampling frequency and re-sampling at least a portion of the sensor data based on the adjusted sampling frequency;
instructing the sensor to adjust the transmit frequency in response to determining that the performance of the machine learning model reaches the threshold using the re-sampled data;
formulating an adjustment ratio by dividing the adjusted sampling frequency by the transmit frequency; and
determining an adjustment type based on the adjustment ratio, wherein the adjustment type is determined by the following steps:
determining that the adjustment type is an upsample adjustment by detecting that the transmit frequency is an integer multiple of the adjusted sampling frequency and in response increasing the sampling frequency; or that the adjustment type is one of a fractional upsample or a fractional downsample adjustment based on the detecting that the transmit frequency is not an integer multiple of the adjusted sampling frequency;
generating a training data set by sampling the set of sensor data based on the adjusted or increased sampling frequency and zero stuffing at least a portion of the training data set where the transmit frequency is not an integer multiple of the adjusted sampling frequency;
filtering the training data set using a finite impulse response (FIR) interpolator on the training data set; and
training the machine learning using the training data set.