Neural Network Parameter Updates via Encoded Input Tokens
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing AI devices often lack network connectivity or sufficient bandwidth to receive neural network (NN) updates, making it difficult to improve the adaptability and long-term performance of deployed models.
Innovation Solution
Implementing a neural network model that can detect encoded messages within input tokens, allowing for on-device updates of parameters using the same sensing modality, eliminating the need for separate processes or additional hardware.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If neural network parameters are updated via network connectivity, then model performance improvement is achieved, but device complexity and connectivity requirements increase
Solution Approach 1:
The patent introduces an intermediary mechanism where encoded messages containing neural network parameter updates are embedded within existing input tokens. This allows the update process to occur through the normal input stream without requiring separate network connectivity or additional communication channels, thus resolving the contradiction between model performance improvement and device complexity
Solution Approach 2:
The patent makes the input token serve multiple functions: it simultaneously carries both the original input data for inference and the encoded neural network parameter updates. This multi-functionality eliminates the need for separate update transmission channels and reduces overall system complexity while maintaining model performance improvement capabilities
2Device complexity
If neural network parameters are hardcoded onto silicon chips, then device simplicity is maintained, but adaptability and long-term performance improvement are limited
Solution Approach 1:
The patent transforms the static hardcoded parameters into dynamic, updatable parameters by encoding updates within input tokens. This allows the neural network to adapt its parameters in real-time based on incoming data without requiring physical hardware changes, thus maintaining hardware simplicity while achieving dynamic adaptability and long-term performance improvement
Solution Approach 2:
The patent enables parameter changes without hardware modification by embedding updated parameter values within the input token data stream. This allows the same physical hardware to accommodate evolving model requirements through software-based parameter updates, resolving the contradiction between hardware simplicity and model adaptability
3Adaptability or versatility
If separate update processes are implemented, then parameter updating capability is achieved, but device complexity and processing overhead increase
Solution Approach 1:
The patent merges the parameter update process with the normal inference processing by encoding updates within input tokens. This consolidation eliminates separate update processes and reduces processing overhead, as the same computational infrastructure handles both inference and parameter updates simultaneously without requiring additional dedicated processing steps
Data Source
AI summary
This disclosure provides methods, devices, and systems for machine learning. The present implementations more specifically relate to systems and techniques for updating neural network (NN) parameters via encoded messages. An input device may implement a NN model trained to perform inferencing on input tokens received via one or more sensors of the input device. In some aspects, the input device receives a first input token via the one or more sensors, determines that the first input token includes an encoded message, extracts NN information from the encoded message, and updates one or more parameters of the NN model based on the extracted NN information. In some other aspects, the input device receives a second input token via the one or more sensors, determines that the second input token does not include an encoded message, and performs an inferencing operation on the second input token based on the updated NN model.