Search

Deep learning (LeCun et al. in Nature 521(7553):436–444, 2015) workloads are common in today's production clusters due to the proliferation of deep learning driven AI (Russell and Norvig, Artificial intelligence: a modern approach, Pearson Education, Upper Saddle River, 2003) services. As deep learning workloads and GPUs all grow more heterogeneous, efficient and fair resource scheduling are the key to the maximal performance of a deep learning cluster. However, existing cluster schedulers are largely not tailored to deep learning jobs, and typically specifying a fixed amount of resources for each job, prohibiting high resource efficiency and job performance. Thus we reconsider the problem in this paper, and to resolve them we proposes MixTran, which works fairly and efficiently for mixed deep learning workloads on heterogeneous GPU clusters. At first, MixTran abstracts heterogeneous GPU resources and distributes them fairly to users in the form of virtual tickets. Then, MixTran transfers uniform requests into a global optimization model, which can make efficient use of GPUs and satisfy the quantified resource requests, heterogeneous node constraints and user fairness constraints. Finally, MixTran conducts greedy resource trading according to the second trader price, which can benefit both users. In the evaluation, we prove MixTran can greatly decrease total execution time of deep learning workloads than traditional scheduler up to 30%–50% and maintain the fairness of multiple users. [ABSTRACT FROM AUTHOR]

Fairness has become a critical value online, and the latest studies consider it in many problems. In recommender systems, fairness is important since the visibility of items is controlled by systems. Previous fairness-aware recommender systems assume that sufficient relationship data between users and items are available. However, it is common that new users and items are frequently introduced, and they have no relationship data yet. In this paper, we study recommendation methods to enhance fairness in a cold-start state. Fairness is more significant when the preference of a user or the popularity of an item is unknown. We propose a meta-learning-based cold-start recommendation framework called FaRM to alleviate the unfairness of recommendations. The proposed framework consists of three steps. We first propose a fairness-aware meta-path generation method to eliminate bias in sensitive attributes. In addition, we construct fairness-aware user representations through the meta-path aggregation approach. Then, we propose a novel fairness objective function and introduce a joint learning method to minimize the trade-off between relevancy and fairness. In extensive experiments with various cold-start scenarios, it is shown that FaRM is significantly superior in fairness performance while preserving relevance accuracy over previous work. [ABSTRACT FROM AUTHOR]

Abstract Fairness has become a critical value online, and the latest studies consider it in many problems. In recommender systems, fairness is important since the visibility of items is controlled by systems. Previous fairness-aware recommender systems assume that sufficient relationship data between users and items are available. However, it is common that new users and items are frequently introduced, and they have no relationship data yet. In this paper, we study recommendation methods to enhance fairness in a cold-start state. Fairness is more significant when the preference of a user or the popularity of an item is unknown. We propose a meta-learning-based cold-start recommendation framework called FaRM to alleviate the unfairness of recommendations. The proposed framework consists of three steps. We first propose a fairness-aware meta-path generation method to eliminate bias in sensitive attributes. In addition, we construct fairness-aware user representations through the meta-path aggregation approach. Then, we propose a novel fairness objective function and introduce a joint learning method to minimize the trade-off between relevancy and fairness. In extensive experiments with various cold-start scenarios, it is shown that FaRM is significantly superior in fairness performance while preserving relevance accuracy over previous work.

This article reviews the theory of fairness in AI-from machine learning to federated learning, where the constraints on precision AI fairness and perspective solutions are also discussed. For a reliable and quantitative evaluation of AI fairness, many associated concepts have been proposed, formulated and classified. However, the inexplicability of machine learning systems makes it almost impossible to include all necessary details in the modelling stage to ensure fairness. The privacy worries induce the data unfairness and hence, the biases in the datasets for evaluating AI fairness are unavoidable. The imbalance between algorithms' utility and humanization has further reinforced such worries. Even for federated learning systems, these constraints on precisionAI fairness still exist.Aperspective solution is to reconcile the federated learning processes and reduce biases and imbalances accordingly. [ABSTRACT FROM AUTHOR]

Breast cancer is the most common cancer affecting women globally. Despite the significant impact of deep learning models on breast cancer diagnosis and treatment, achieving fairness or equitable outcomes across diverse populations remains a challenge when some demographic groups are underrepresented in the training data. We quantified the bias of models trained to predict breast cancer stage from a dataset consisting of 1000 biopsies from 842 patients provided by AIM-Ahead (Artificial Intelligence/Machine Learning Consortium to Advance Health Equity and Researcher Diversity). Notably, the majority of data (over 70%) were from White patients. We found that prior to post-processing adjustments, all deep learning models we trained consistently performed better for White patients than for non-White patients. After model calibration, we observed mixed results, with only some models demonstrating improved performance. This work provides a case study of bias in breast cancer medical imaging models and highlights the challenges in using post-processing to attempt to achieve fairness. [ABSTRACT FROM AUTHOR]

Artificial Intelligence (AI) models for medical diagnosis often face challenges of generalizability and fairness. We highlighted the algorithmic unfairness in a large thyroid ultrasound dataset with significant diagnostic performance disparities across subgroups linked causally to sample size imbalances. To address this, we introduced the Quasi-Pareto Improvement (QPI) approach and a deep learning implementation (QP-Net) combining multi-task learning and domain adaptation to improve model performance among disadvantaged subgroups without compromising overall population performance. On the thyroid ultrasound dataset, our method significantly mitigated the area under curve (AUC) disparity for three less-prevalent subgroups by 0.213, 0.112, and 0.173 while maintaining the AUC for dominant subgroups; we also further confirmed the generalizability of our approach on two public datasets: the ISIC2019 skin disease dataset and the CheXpert chest radiograph dataset. Here we show the QPI approach to be widely applicable in promoting AI for equitable healthcare outcomes. Artificial Intelligence (AI) models for medical diagnosis often face challenges of generalizability and fairness. Here, the authors show that the Quasi-Pareto Improvement approach is widely applicable to improving AI models among less-prevalent subgroups, promoting equitable healthcare outcomes. [ABSTRACT FROM AUTHOR]