Search

Mitigating bias in machine learning models is a critical endeavor for ensuring fairness and equity. In this paper, we propose a novel approach to address bias by leveraging pixel image attributions to identify and regularize regions of images containing significant information about bias attributes. Our method utilizes a model-agnostic approach to extract pixel attributions by employing a convolutional neural network (CNN) classifier trained on small image patches. By training the classifier to predict a property of the entire image using only a single patch, we achieve region-based attributions that provide insights into the distribution of important information across the image. We propose utilizing these attributions to introduce targeted noise into datasets with confounding attributes that bias the data, thereby constraining neural networks from learning these biases and emphasizing the primary attributes. Our approach demonstrates its efficacy in enabling the training of unbiased classifiers on heavily biased datasets., Comment: Accepted to the 2024 IEEE CVPR Workshop on Fair, Data-efficient, and Trusted Computer Vision

Camera sensors are increasingly being combined with machine learning to perform various tasks such as intelligent surveillance. Due to its computational complexity, most of these machine learning algorithms are offloaded to the cloud for processing. However, users are increasingly concerned about privacy issues such as function creep and malicious usage by third-party cloud providers. To alleviate this, we propose an edge-based filtering stage that removes privacy-sensitive attributes before the sensor data are transmitted to the cloud. We use state-of-the-art image manipulation techniques that leverage disentangled representations to achieve privacy filtering. We define opt-in and opt-out filter operations and evaluate their effectiveness for filtering private attributes from face images. Additionally, we examine the effect of naturally occurring correlations and residual information on filtering. We find the results promising and believe this elicits further research on how image manipulation can be used for privacy preservation., Comment: Accepted to the MLCS workshop at ECML PKDD 2022

Zinc tin oxide (Zn1-xSnxOy) has been proposed as an alternative buffer layer material to the toxic, and light narrow-bandgap CdS layer in CuIn1-x,GaxSe2 thin film solar cell modules. In this present study, synchrotron-based soft X-ray absorption and emission spectroscopies have been employed to probe the densities of states of intrinsic ZnO, Zn1-xSnxOy and SnOx thin films grown by atomic layer deposition. A distinct variation in the bandgap is observed with increasing Sn concentration, which has been confirmed independently by combined ellipsometry-reflectometry measurements. These data correlate directly to the open circuit potentials of corresponding solar cells, indicating that the buffer layer composition is associated with a modification of the band discontinuity at the CIGS interface. Resonantly excited emission spectra, which express the admixture of unoccupied O 2p with Zn 3d, 4s, and 4p states, reveal a strong suppression in the hybridization between the O 2p conduction band and the Zn 3d valence band with increasing Sn concentration.

The electronic structure change during the reversible Li-ion storage reaction in a bimetallic MnFe-Prussian blue analogue (Li(x)K(0.14)Mn(1.43)[Fe(CN)(6)] center dot 6H(2)O) was investigated by soft x-ray absorption spectroscopy. The Mn L(2,3)-edgespectra revealed the unchanged Mn(2+) high-spin state regardless of Li-ion concentration (x). On the other hand, the Fe L(2,3)-edge spectra clearly revealed a reversible redox behavior as Fe(3+) Fe(2+) states with Li-ion insertion/extraction. Experimental findings suggested strong metal-to-ligand charge transfer in conjunction with the ligand-to-metal one. The resulting charge delocalization between the Fe and CN is considered to contribute to the high reversibility of the Li-ion storage process.