Machine-Learned Discretization Level Reduction for Tensor Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for binarizing tensor data, such as images, often fail to maintain visual integrity and detail, leading to uninterpretable or noisy output, especially when reducing high-color images to black and white, and face challenges in training machine learning models due to the lack of suitable training data.
Innovation Solution
A computer-implemented method using a machine-learned discretization level reduction model that progressively reduces the number of discretization levels in tensor data through a series of level reduction layers, with a discretized activation function and a color bypass network, allowing for the reconstruction of input data to train the model and improve visual representation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If conventional binarization methods are used to reduce discretization levels, then data compression is achieved, but visual integrity and detail are lost
Solution Approach 1:
The model performs preliminary learning during training by reconstructing high-resolution images from low-resolution inputs, enabling it to anticipate and preserve important visual features before the actual binarization process occurs during inference
Solution Approach 2:
The reconstruction network creates a copy of the original high-resolution image from the binarized representation, allowing the model to learn the mapping between compressed and original data while preserving visual integrity in the reconstructed output
2Loss of energy
If discretization levels are reduced to compress images, then storage efficiency improves, but color information and boundaries are lost
Solution Approach 1:
The reconstruction network provides feedback during training by comparing the reconstructed image with the original, allowing the model to learn which color and boundary information is most critical to preserve during discretization level reduction
Solution Approach 2:
The model learns to change the discretization parameters dynamically, using continuous relaxation techniques that allow gradients to flow through the discretization operation, enabling optimization of color and boundary preservation alongside compression
3Extent of automation
If machine learning models are trained for binarization, then automated processing improves, but training data scarcity becomes a problem
Solution Approach 1:
The model is self-supervised, using the original high-resolution images themselves as training targets for the reconstruction network, eliminating the need for external annotated training data while enabling automated binarization processing
Data Source
AI summary
A computer-implemented method for providing level-reduced tensor data having improved representation of information can include obtaining input tensor data, providing the input tensor data as input to a machine-learned discretization level reduction model configured to receive tensor data having a number of discretization levels and produce, in response to receiving the tensor data, level-reduced tensor data having a reduced number of discretization levels, and obtaining, from the machine-learned discretization level reduction model, the level-reduced tensor data. The machine-learned discretization level reduction model is trained using reconstructed input tensor data generated using an output of the machine-learned discretization level reduction model. The machine-learned discretization level reduction model can include one or more level reduction layers configured to receive input having a first number of discretization levels and to provide a layer output having a reduced a number of discretization levels.