Deep neural networks trained on biased data often inadvertently learn unintended inference rules, particularly when labels are strongly correlated with biased features. Existing bias mitigation methods typically involve either a) predefining bias types and enforcing them as prior knowledge or b) reweighting training samples to emphasize bias-conflicting samples over bias-aligned samples. However, both strategies address bias indirectly in the feature or sample space, with no control over learned weights, making it difficult to control the bias propagation across different layers. Based on this observation, we introduce a novel approach to address bias directly in the model's parameter space, preventing its propagation across layers. Our method involves training two models: a bias model for biased features and a debias model for unbiased details, guided by the bias model. We enforce dissimilarity in the debias model's later layers and similarity in its initial layers with the bias model, ensuring it learns unbiased low-level features without adopting biased high-level abstractions. By incorporating this explicit constraint during training, our approach shows enhanced classification accuracy and debiasing effectiveness across various synthetic and real-world datasets of different sizes. Moreover, the proposed method demonstrates robustness across different bias types and percentages of biased samples in the training data. The code is available at: https://visdomlab.github.io/CosFairNet/

Learning discriminative features is crucial for various robotic applications such as object detection and classification. In this paper, we present a general framework for the analysis of the discriminative properties of haptic signals. Our focus is on two crucial components of a robotic perception system: discriminative feature extraction and metric-based feature transformation to enhance the separability of haptic signals in the projected space. We propose a set of hand-crafted haptic features (generated only from acceleration data), which enables discrimination of real-world textures. Since the Euclidean space does not reflect the underlying pattern in the data, we propose to learn an appropriate transformation function to project the feature onto the new space and apply different pattern recognition algorithms for texture classification and discrimination tasks. Unlike other existing methods, we use a triplet-based method for improved discrimination in the embedded space. We further demonstrate how to build a haptic vocabulary by selecting a compact set of the most distinct and representative signals in the embedded space. The experimental results show that the proposed features augmented with learned embedding improves the performance of semantic discrimination tasks such as classification and clustering and outperforms the related state-of-the-art.