USPTO Patent Grant: Deep Learning Enhanced Garbage Collection by HPE

ChangeBridge: Patent Grants - AI & Computing (G06N)

Published March 24th, 2026

Detected March 24th, 2026

Summary

The USPTO has granted Hewlett Packard Enterprise Development LP a patent (US12585937B2) for systems and methods related to deep learning enhanced garbage collection. The patent details a process for forecasting data deletes to optimize garbage collection scheduling in data arrays.

View original document View source feed page

What changed

The United States Patent and Trademark Office (USPTO) has issued patent US12585937B2 to Hewlett Packard Enterprise Development LP. This patent covers novel systems and methods for enhancing garbage collection in data storage arrays using deep learning techniques. The disclosed technology involves analyzing time series data of writes and deletes, encoding it into tensors, identifying patterns, forecasting future deletes, and then scheduling garbage collection algorithms based on these predictions to optimize array performance.

This patent grant is a non-binding notice of intellectual property and does not impose new regulatory obligations on companies. However, it signifies innovation in AI-driven data management and could influence future product development and industry standards in computing and data storage. Companies operating in this space should be aware of this patented technology, particularly if developing similar AI-based garbage collection solutions.

Source document (simplified)

← USPTO Patent Grants

Systems and methods for deep learning enhanced garbage collection

Grant US12585937B2 Kind: B2 Mar 24, 2026

Assignee

Hewlett Packard Enterprise Development LP

Inventors

Sajjit Thampy, Krishna Sirisha Motamarry, Gyan Bhal

Abstract

Examples disclosed herein relate to systems and methods for deep learning enhanced garbage collection. Disclosed methods may include receiving, at a controller of an array, a time series dataset including a number of writes and a number of deletes; generating, by the controller, a tensor by encoding the time series data using a plurality of frequencies; determining a pattern in the time series data; generating a forecast of deletes based on the pattern; determining, based on the forecast, a number of cumulative deletes to the array at a predetermined time; at the predetermined time, comparing the forecasted number of cumulative deletes with a number of actual deletes to the array; and based on the comparison, scheduling a garbage collection (GC) algorithm to run on the array.