Enhancing The Discovery Of Neural Representations: Integrating Task-Relevant Dimensionality Reduction And Domain Adaptation