16 research outputs found
Guess, Learn, Repeat: Intelligent Learning System with Synthetic and Counterfactual Training in a GeoGuessr-Inspired Classification Task
No full text53805389Training novices by experts is often costly and time-consuming. Alternatively, learning systems offer a scalable and automated alternative. However, learning systems offer another, yet underexplored advantage, over training with experts: Analyzing novices and providing personalized training. This study explores the use of synthetically generated images to improve novice image classification skills in a GeoGuessr-inspired classification task. By leveraging a counterfactual-based approach and synthetically generated personalized training data, we aim to enhance individual learning. In a controlled experiment where participants classify Google Street View images from four different cities, we compare the impact of personalized synthetic images against randomly assigned ones. Our findings indicate that personalized training improves classification accuracy, underscoring the potential of intelligent learning. These results highlight a promising direction for integrating synthetic data into adaptive training environments in game-like settings, paving the way for effective and personalized intelligent learning systems
Guess, Learn, Repeat: Intelligent Learning System with Synthetic and Counterfactual Training in a GeoGuessr-Inspired Classification Task
No full textGuess, Learn, Repeat: Intelligent Learning System with Synthetic and Counterfactual Training in a GeoGuessr-Inspired Classification Task
No full textTraining novices by experts is often costly and time-consuming. Alternatively, learning systems offer a scalable and automated alternative. However, learning systems offer another, yet underexplored advantage, over training with experts: Analyzing novices and providing personalized training. This study explores the use of synthetically generated images to improve novice image classification skills in a GeoGuessr-inspired classification task. By leveraging a counterfactual-based approach and synthetically generated personalized training data, we aim to enhance individual learning. In a controlled experiment where participants classify Google Street View images from four different cities, we compare the impact of personalized synthetic images against randomly assigned ones. Our findings indicate that personalized training improves classification accuracy, underscoring the potential of intelligent learning. These results highlight a promising direction for integrating synthetic data into adaptive training environments in game-like settings, paving the way for effective and personalized intelligent learning systems
Guess, Learn, Repeat: Intelligent Learning System with Synthetic and Counterfactual Training in a GeoGuessr-Inspired Classification Task
No full textTraining novices by experts is often costly and time-consuming. Alternatively, learning systems offer a scalable and automated alternative. However, learning systems offer another, yet underexplored advantage, over training with experts: Analyzing novices and providing personalized training. This study explores the use of synthetically generated images to improve novice image classification skills in a GeoGuessr-inspired classification task. By leveraging a counterfactual-based approach and synthetically generated personalized training data, we aim to enhance individual learning. In a controlled experiment where participants classify Google Street View images from four different cities, we compare the impact of personalized synthetic images against randomly assigned ones. Our findings indicate that personalized training improves classification accuracy, underscoring the potential of intelligent learning. These results highlight a promising direction for integrating synthetic data into adaptive training environments in game-like settings, paving the way for effective and personalized intelligent learning systems
A Multivocal Literature Review on Privacy and Fairness in Federated Learning
No full textFederated Learning presents a way to revolutionize AI applications by eliminating the necessity for data sharing. Yet, research has shown that information can still be extracted during training, making additional privacy-preserving measures such as differential privacy imperative. To implement real-world federated learning applications, fairness, ranging from a fair distribution of performance to non-discriminative behavior, must be considered. Particularly in high-risk applications (e.g. healthcare), avoiding the repetition of past discriminatory errors is paramount. As recent research has demonstrated an inherent tension between privacy and fairness, we conduct a multivocal literature review to examine the current methods to integrate privacy and fairness in federated learning. Our analyses illustrate that the relationship between privacy and fairness has been neglected, posing a critical risk for real-world applications. We highlight the need to explore the relationship between privacy, fairness, and performance, advocating for the creation of integrated federated learning frameworks
Designing CausalML for Effective Human–AI Collaboration
No full textThis research project investigates how causal information provided by AI systems influences human-AI collaboration. Specifically, the focus lies on how a) the type of AI advice (causal vs. non-causal) affects trust and reliance, and b) how an introduction on correlation and causality influences trust and reliance but also moderates the effect of the AI advice. Further, we examine how individual background knowledge about AI and statistics may moderate the effect.
Therefore, our central research questions are:
(1) How does the type of AI advice (causal vs. non-causal information) influence decision-makers’ trust in and reliance on the AI system?
(2) How does the decision-makers’ background knowledge about AI and statistics moderate the effect of the type of AI advice on trust in and reliance on the AI system?
(3) How does an introduction on correlation and causality influence trust in and reliance on the AI system, and how does it moderate the effect of the type of AI advice on trust and reliance?
(4) How does the decision-makers’ background knowledge about AI and statistics moderate the effect of an introduction on correlation and causality on trust in and reliance on the AI system?
To this goal, we use a randomized controlled trial (RCT) in a 2x2 between-subjects study where participants decide whether a job training should be applied to applicants after seeing profiles containing socio-demographic information.
Participants are supported by an AI system. The type of AI advice is the first factor of our design, with two levels: causal vs. non-causal information. Whether participants receive an introduction on correlation and causality before the study is the second factor, with two levels: receiving an introduction vs. receiving no introduction. The introduction explains the distinction between correlation and causality and contrasts standard with causal machine learning models.
In the study, participants are randomly assigned to one of the four conditions. Participants in the introduction condition self-report their statistical literacy, then receive the introduction at the beginning of the study, which is identical across the types of AI advice conditions. In every condition, participants sequentially evaluate ten profiles of potential applicants to receive job training. For each profile, they make an initial decision about whether to assign the job training based on whether they assume a job training will lead to a higher future income than no training. Further, participants are made aware of the limited availability of job trainings. After making their decisions sequentially, participants are presented with an overview of all decisions and are allowed to make adjustments. Afterwards, they rate their confidence. They are then supported by binary AI advice (Provide job training/Do not provide job training) and re-evaluate the same profiles sequentially, with their original decisions displayed alongside the AI’s suggestions. In the non-causal condition, the AI advice additionally contains the expected future income when receiving a job training, as predicted by a traditional ML model. In the causal conditions, it shows the predicted future income with and without job training, as well as the individual treatment effect predicted by a causal ML model. Again, participants are provided with an overview of their decisions and are allowed to make adjustments. Afterwards, they rate their confidence again.
Subsequently, we collect demographics, self-reported AI and statistical competencies, validated trust scales, and perceived fairness
Collafuse: Navigating Limited Resources and Privacy in Collaborative Generative AI
No full textIn the landscape of generative artificial intelligence, diffusion models present challenges for socio-technical systems in data requirements and privacy. Traditional approaches like federated learning distribute the learning process but strain individual clients, especially with constrained resources. In response to these challenges, we introduce \textsc{CollaFuse}, a novel framework inspired by split learning. Tailored for efficient and collaborative use of denoising diffusion probabilistic models, \textsc{CollaFuse} enables shared server training and inference, alleviating client computational burdens. This is achieved by retaining data and computationally inexpensive GPU processes locally at each client while outsourcing the computationally expensive processes to the shared server. Demonstrated in a healthcare context, \textsc{CollaFuse} enhances privacy by reducing the need for sensitive information sharing. These capabilities hold the potential to impact various application areas, such as edge computing, healthcare research, or autonomous driving. In essence, our work advances distributed machine learning, shaping the future of collaborative GenAI networks
CollaFuse: Navigating Limited Resources and Privacy in Collaborative Generative AI
No full textIn the landscape of generative artificial intelligence, diffusion models present challenges for socio-technical systems in data requirements and privacy. Traditional approaches like federated learning distribute the learning process but strain individual clients, especially with constrained resources. In response to these challenges, we introduce \textsc{CollaFuse}, a novel framework inspired by split learning. Tailored for efficient and collaborative use of denoising diffusion probabilistic models, \textsc{CollaFuse} enables shared server training and inference, alleviating client computational burdens. This is achieved by retaining data and computationally inexpensive GPU processes locally at each client while outsourcing the computationally expensive processes to the shared server. Demonstrated in a healthcare context, \textsc{CollaFuse} enhances privacy by reducing the need for sensitive information sharing. These capabilities hold the potential to impact various application areas, such as edge computing, healthcare research, or autonomous driving. In essence, our work advances distributed machine learning, shaping the future of collaborative GenAI networks
It's only fair when I think it's fair: How Gender Bias Alignment Undermines Distributive Fairness in Human-AI Collaboration
No full text12611274Human-AI collaboration is increasingly relevant in consequential areas where AI recommendations support human discretion. However, human-AI teams’ effectiveness, capability, and fairness highly depend on human perceptions of AI. Positive fairness perceptions have been shown to foster trust and acceptance of AI recommendations. Yet, work on confirmation bias highlights that humans selectively adhere to AI recommendations that align with their expectations and beliefs - despite not being necessarily correct or fair. This raises the question whether confirmation bias also transfers to the alignment of gender bias between human and AI decisions. In our study, we examine how gender bias alignment influences fairness perceptions and reliance. The results of a 2x2 between-subject study highlight the connection between gender bias alignment, fairness perceptions, and reliance, demonstrating that merely constructing a "formally fair" AI system is insufficient for optimal human-AI collaboration; ultimately, AI recommendations will likely be overridden if biases do not align
