Video Decoder Neural Inference Reliability Using Analog Metadata
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Analog devices used in neural network inference processes for video coding produce non-bit-exact output results, leading to unreproducible outcomes, which is unacceptable in video codecs where identical inputs should yield identical outputs.
Innovation Solution
Implement a method to obtain metadata representing the reliability of the neural network inference process by an analog device, using checksums or statistical parameters to ensure output consistency, and decode video data based on these metadata to achieve reproducible results.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Use of energy by moving object
If analog devices are used to implement neural network inference processes, then energy consumption is reduced, but output reproducibility deteriorates
Solution Approach 1:
The encoder pre-computes the neural network inference output using a digital device and stores the result in a buffer before transmission. This preliminary action ensures that the reference output is deterministic and reproducible, while the analog device at the decoder can operate with lower energy consumption without compromising reproducibility, as it only needs to verify against the pre-stored reference.
Solution Approach 2:
A buffer storing pre-computed neural network output acts as an intermediary between the encoder and decoder. The buffer provides the deterministic reference output that the decoder compares against its analog device output, mediating between the need for reproducibility and the energy efficiency of analog computation.
2Reliability
If digital devices are used to ensure output reproducibility, then reliability is improved, but energy consumption increases
Solution Approach 1:
The digital device performs the neural network inference computation in advance during encoding, storing the deterministic output in a buffer. This preliminary computation ensures reproducibility is established upfront, allowing the energy-intensive digital operation to occur only once rather than repeatedly at the decoder.
Solution Approach 2:
The deterministic output from the digital device is copied into a buffer and transmitted to the decoder. This copying mechanism allows the decoder to use a stored reference rather than performing its own energy-intensive digital computation, reducing overall energy consumption while maintaining reproducibility.
3Productivity
If analog devices are used for neural network inference, then processing speed is improved, but manufacturing precision deteriorates
Solution Approach 1:
The encoder pre-computes the exact neural network output using a digital device and stores it in a buffer before transmission. This preliminary action establishes the precise reference output that the analog device at the decoder will compare against, allowing the analog device to operate faster without compromising output exactness.
Solution Approach 2:
The decoder compares the analog device output against the pre-stored reference output from the buffer, creating a feedback mechanism. This feedback allows the system to verify whether the fast analog computation produced an acceptable result, and if not, to request re-computation or correction, thereby maintaining manufacturing precision despite the use of faster analog hardware.
Data Source
AI summary
A method comprising obtaining (611) video data comprising meta-data representative of a level of reliability of a neural network inference process implemented by an analog device, the neural network inference process being applied to decode the video data, an analog device being an electronic circuitry unable to ensure a repeatability of an output result of said electronic circuitry when said electronic circuitry receives same input data; and, decoding (612) the video data, the implementation of the neural network inference process by the analog device depending on the metadata.