Bias mitigationClient selection strategies have become a widely adopted approach in recent years within the studies on Federated Learning (FL). This strategy aims to handle the communication efficiency problem between the server and clients. However, due to certain differences in data distribution among...
Semi decentralized algorithms may improve convergence rates and reduce global update bias in federated learning applications [22]. The FL-EOCD approach also provides energy efficient decentralized aggregation in device to device networks [23]. Further research proposes a hybrid strategy integrating device ...
The ΔD and PQD of FairerOPTH for each disease are shown in Fig. 3c, d. Compared with the baseline model, FairerOPTH has smaller ΔD values for 38 diseases, which means that FairerOPTH is fairer when considering the sex of patients, showing less sex bias, especially for diagnosing ...
a malicious participant in DFL Information Description Model parameters Topology Roles Metrics Activity periods Model architecture Communication patterns Each layer li in a model M with n layers has weight wi 2 RdiÂdiÀ1 and bias bi 2 Rdi , where di is the number of neurons in layer i. ...
We prove the bias-variance bound used in FairDP in Appendix A. We can now state our main theorem. Theorem 1 Suppose the function f is L-Lipschitz smooth, for the t-th iteration of differentially private SGD with learning rate μt, batch size n, clipping threshold parameter C and noise ...
We study Sn and Qn by means of their influence functions, their bias curves (for implosion as well as explosion), and their finite-sample performance. Their behavior is also compared at non-Gaussian models, including the negative exponential model where Sn has a lower gross-error sensitivity ...
and remark that an LLM can move away from the ideal performance range due to bias human feedback. State-of-the-art solutions typically involve utilizing a KL-Divergence based brute-force approach for declining upgrades to the LLM model in a mixed-trust scenario. However, such an approach can...
In this paper, we study the issue of popularity bias in recommendation systems under the federated learning framework. First, we quantitatively analyze the popularity bias in federated recommendation models, demonstrating the presence of strong popularity bias in their recommendation results. Secondly, ...
medical image analysis; digital ageism; bias mitigation; skin lesion detection; artificial intelligence1. Introduction In recent years, the success of deep learning models across a wide range of application domains has spurred the creation of numerous large datasets. However, these datasets often ...