Neural Network Tensor Concatenation for Low-Complexity Compression
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Trained neural networks have a large size due to numerous parameters and weights, necessitating efficient compression methods to reduce their size without compromising performance.
Innovation Solution
A method for encoding and decoding tensors by swapping axes to create concatenable tensors, reshaping dimensions, concatenating them, and generating syntax structures for compression, along with a bitstream combining these structures and compressed tensors.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If neural networks are compressed to reduce size, then storage and transmission efficiency improve, but decoding complexity increases due to axis swapping and reshaping operations
Solution Approach 1:
The patent applies preliminary action by performing axis swapping and reshaping operations during the encoding phase before compression. The encoder reorganizes tensor dimensions and swaps axes to create a standardized format, so that the decoder only needs to perform simple inverse operations. This pre-processing eliminates the need for complex decoding logic, resolving the contradiction between compression and decoding complexity.
Solution Approach 2:
The patent uses copying by storing axis swapping information and dimension metadata in syntax structures within the bitstream. Instead of performing complex computations during decoding, the decoder reads the pre-stored swapping instructions and applies simple transformations. This copies the complexity information into the bitstream, allowing the decoder to remain computationally simple while still achieving accurate reconstruction.
2Productivity
If tensors are reshaped and axes are swapped to enable concatenation, then compression efficiency improves, but encoding complexity increases
Solution Approach 1:
The patent applies parameter changes by systematically transforming tensor dimensions through controlled axis swapping and reshaping operations. The encoder changes the dimensional parameters of tensors to match a standardized format, enabling efficient concatenation along specific axes. These parameter transformations are governed by clear rules stored in syntax structures, making the increased encoding complexity manageable and systematic rather than arbitrary.
3Quantity of substance
If multiple tensors are concatenated around a specific axis, then compression ratio improves, but dimension compatibility requirements worsen
Solution Approach 1:
The patent applies segmentation by dividing the tensor concatenation process into distinct stages: first identifying compatible axes through dimension analysis, then performing axis swapping to align dimensions, and finally concatenating along the specified axis. This segmented approach allows the system to handle dimension compatibility requirements systematically, checking and adjusting each dimension independently before concatenation, thereby enabling higher compression ratios through multi-tensor concatenation.
Data Source
AI summary
The embodiments relate to a method for encoding two or more tensors. The method comprises processing the two or more tensors having respective dimensions so that the dimensions of said two or more sensors have the same number (510); identifying which axis of each individual tensor is swappable to result in concatenable tensors around an axis of concatenation (520); reshaping the tensors so that the dimensions are modified based on the swapped axis (530); concatenating the tensors around the axis of concatenation to result in concatenated tensor (540); compressing the concatenated tensor (550); generating syntax structures for carrying concatenation and axis swapping information (560); and generating a bitstream by combining the syntax structures and the compressed concatenated tensor (570). The embodiments also relate to a method for decoding, and to apparatuses for implementing the methods.