Deep Memory Network External Memory Controller
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing intelligent services models require extensive pre-training and retraining, struggle to remember user interaction history, and fail to adapt to users' evolving needs over time, limiting their ability to provide personalized and dynamic services.
Innovation Solution
A deep memory network system that incorporates an external memory to store and update user data and feedback, allowing the model to continuously adapt to new information and user dynamics without the need for retraining, using a controller to manage read and write operations within the external memory.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If existing intelligent service models are used, then they can provide basic services, but they require extensive pre-training and retraining which is costly and difficult
Solution Approach 1:
The model is segmented into a core pre-trained component and a separate external memory component. The external memory is further divided into key memory and value memory slots that can be independently updated. This segmentation allows the core model to remain fixed while only the memory components require adaptation, significantly reducing retraining overhead.
Solution Approach 2:
The system performs preliminary action by pre-training the core model once on general tasks, then equipping it with an external memory structure that can be rapidly adapted to specific users and tasks without retraining the core model. The memory slots are pre-configured to accept new information, enabling quick adaptation to new scenarios.
2Adaptability or versatility
If existing models collect user history data, then they can attempt personalization, but they fail to remember and adapt to users' evolving needs over time
Solution Approach 1:
An external memory structure acts as an intermediary between the core model and user interaction history. The memory comprises key slots that store user identifiers and value slots that store interaction histories and preferences. This intermediary enables the model to access and adapt to user-specific information without modifying the core model architecture, achieving continuous adaptation to evolving user needs.
3Measurement precision
If models require many interactions to provide personalized services, then they can learn user preferences, but the process is inefficient and time-consuming
Solution Approach 1:
The system uses copying by storing user profiles, interaction histories, and preferences as copyable data in the external memory value slots. When a user interacts with the system, their information is copied into the memory slots, allowing the model to immediately access personalized information without requiring extensive relearning. This copying mechanism enables rapid personalization from fewer interactions.
Data Source
AI summary
An electronic device including a deep memory model includes at least one memory and at least one processor coupled to the at least one memory. The at least one processor is configured to receive input data to the deep memory model. The at least one processor is also configured to extract a history state of an external memory coupled to the deep memory model based on the input data. The at least one processor is further configured to update the history state of the external memory based on the input data. In addition, the at least one processor is configured to output a prediction based on the extracted history state of the external memory.