Publication

Melanoma Detection through Combining Reinforcement Learning, Generative Adversarial Network, and Bayesian Optimization

2026

Abstract

Melanoma, a highly aggressive form of skin cancer, is primarily driven by DNA alterations often linked to environmental factors such as ultraviolet radiation. Addressing the need for improved early detection, this study tackles the key limitations of current methods, which frequently employ convolutional neural networks (CNNs) but struggle with feature selection, class imbalance, hyperparameter tuning, and generalizability. Our strategy leverages dilated convolution (DC) layers trained using reinforcement learning (RL). Unlike other RL-based approaches that handle these challenges in isolation, our method introduces a multi-stage architecture. It integrates RL for feature selection and class balancing. Shapley additive explanations (SHAP) guide feature identification, while augmented rewards for underrepresented classes help mitigate data imbalance. Bayesian optimization hyperband (BOHB) is used for hyperparameter tuning in a unified training process. BOHB combines the predictive strength of Bayesian optimization with the efficiency of hyperband, accelerating model tuning. It also includes an online GAN module for dynamic data augmentation that responds to the evolving output of the RL agent. A novel regularization technique stabilizes GAN training and prevents mode collapse. Importantly, existing RL methods face the challenge of balancing exploration and exploitation. In our RL model, the scope loss function (SLF), integrated with RL, balances exploration and exploitation, thereby ensuring accuracy and generalizability. Collectively, the model jointly tackles four persistent challenges in earlier RL-based approaches: poor exploration–exploitation balance, unstable reward dynamics, static data augmentation, and manual hyperparameter tuning. The model achieved F-measures of 94.3 %, 93.7 %, and 91.5 % on ISIC-2020, HAM10000, and PH2, respectively. This advancement significantly improves early melanoma detection and supports more accurate treatment decisions, contributing valuably to the ongoing effort to combat this lethal cancer.

Official source

https://infoscience.epfl.ch/entities/publication/26a37582-700f-4256-b656-9092682bcfc7

About this result

This page is automatically generated and may contain information that is not correct, complete, up-to-date, or relevant to your search query. The same applies to every other page on this website. Please make sure to verify the information with EPFL's official sources.

Graph Chat

Chat with Graph Search

Ask any question about EPFL courses, lectures, exercises, research, news, etc. or try the example questions below.

DISCLAIMER: The Graph Chatbot is not programmed to provide explicit or categorical answers to your questions. Rather, it transforms your questions into API requests that are distributed across the various IT services officially administered by EPFL. Its purpose is solely to collect and recommend relevant references to content that you can explore to help you answer your questions.

2026

Abstract

Official source

https://infoscience.epfl.ch/entities/publication/26a37582-700f-4256-b656-9092682bcfc7

About this result

Related concepts (32)

Reinforcement learning

Reinforcement learning (RL) is an area of machine learning concerned with how intelligent agents ought to take actions in an environment in order to maximize the notion of cumulative reward. Reinforcement learning is one of three basic machine learning paradigms, alongside supervised learning and unsupervised learning. Reinforcement learning differs from supervised learning in not needing labelled input/output pairs to be presented, and in not needing sub-optimal actions to be explicitly corrected.

Deep reinforcement learning

Deep reinforcement learning (deep RL) is a subfield of machine learning that combines reinforcement learning (RL) and deep learning. RL considers the problem of a computational agent learning to make decisions by trial and error. Deep RL incorporates deep learning into the solution, allowing agents to make decisions from unstructured input data without manual engineering of the state space. Deep RL algorithms are able to take in very large inputs (e.g.

Hyperparameter optimization

In machine learning, hyperparameter optimization or tuning is the problem of choosing a set of optimal hyperparameters for a learning algorithm. A hyperparameter is a parameter whose value is used to control the learning process. By contrast, the values of other parameters (typically node weights) are learned. The same kind of machine learning model can require different constraints, weights or learning rates to generalize different data patterns.

Related publications (32)

Melanoma Detection through Combining Reinforcement Learning, Generative Adversarial Network, and Bayesian Optimization

Chat with Graph Search

On Balanced Representation Learning in Neural Networks

Stable Optimization in Deep Learning: Geometry and Games

Optimizing hepatitis C diagnosis through reinforcement learning feature selection and multi-model machine learning evaluation