Storage Device Model Instantiation Weight Updates
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing storage devices, such as SSDs, face inefficiencies in processing circuitry due to the need to train model instantiations locally, which consumes bandwidth and power, and compounds these issues in systems with multiple storage devices.
Innovation Solution
Implementing a system where a centralized host trains a model instantiation and updates weights for storage devices, allowing these devices to focus on executing operations rather than training, thereby offloading the training task and improving bandwidth and power efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If storage devices train model instantiations locally, then the model can be adapted to local data patterns, but bandwidth and power consumption increase significantly
Solution Approach 1:
The patent extracts the model training function from individual storage devices and centralizes it in a host system. The host trains the model using aggregated data from multiple storage devices, then distributes the trained model weights back to the devices. This separation allows local adaptation without requiring each device to perform computationally intensive training operations locally, thereby reducing power consumption and bandwidth usage while maintaining adaptability.
Solution Approach 2:
The patent merges the training data from multiple storage devices into a centralized host system for unified model training. By combining data resources and computational capabilities at the host level, the system achieves better model generalization and reduces redundant training operations across individual devices, thereby optimizing bandwidth and power efficiency while maintaining adaptability to local patterns.
2Ease of operation
If multiple storage devices train models locally, then each device can operate independently, but system-wide efficiency decreases due to compound resource consumption
Solution Approach 1:
The patent implements preliminary model training at the host level before distributing the trained model to storage devices. This preliminary action ensures that each device receives a pre-optimized model that has already learned from aggregated data patterns, allowing independent operation without requiring subsequent local training. This approach maintains operational independence while eliminating the compound resource consumption that would result from each device training separately.
3Adaptability or versatility
If storage devices perform both training and execution operations, then full functionality is available locally, but access efficiency and power consumption are degraded
Solution Approach 1:
The patent segments the machine learning workflow into distinct phases: model training is performed centrally at the host level, while model execution and inference are performed locally at the storage device level. This segmentation allows each component to specialize in its optimal function - the host handles computationally intensive training operations, while storage devices handle data-intensive execution operations. This division maintains full functionality while improving access efficiency and reducing power consumption by avoiding redundant training operations at each device.
Data Source
AI summary
A system and related method, including memory and processing circuitry, which is to receive multiple weights from a host device to implement an instantiation of a model on the storage device. The instantiation of the model includes multiple weights, and each weight is determined using another instantiation of the model that was trained on the host device. The processing circuitry is then to generate inference information using the instantiation of the model implemented on the storage device. The processing circuitry is further to receive multiple updated weights from the host device and update the instantiation of the model based on the multiple updated weights to implement an updated instantiation of the model on the storage device. The processing circuitry is then to generate updated inference information on the storage device using the updated instantiation of the model based on at least one signal to be processed by the processing circuitry.