⏩ It advances the understanding of how artificial and biological systems differ and lays the foundation for developing vision models that not only perform well but also “see” the world in a more human-like way.

🧩 His thesis proposes a shift from purely performance-driven evaluation toward perceptually informed design principles. By integrating neuroscience, psychophysics, and machine learning.

📙 Dissecting the Effectiveness of Deep Features as Metric of Perceptual Image Quality
🔗

📊 Results from different works reveal that perceptual alignment depends strongly on the statistics of the visual environment and the type of supervision.

📘 Alignment of Color Discrimination in Humans and Image Segmentation Networks
🔗

🔬 The second part addresses human perceptual alignment, developing and applying evaluation protocols inspired by psychophysics to measure the similarity between artificial and human perception.

📗 Why Divisive Normalization Works in Image Segmentation? 
🔗

👁️ However, the inclusion of biologically inspired components does not automatically produce perceptual behavior comparable to human vision.

📕 Neural Networks with Divisive Normalization for Image Segmentation
🔗

🔬 Experiments demonstrate that DN-enhanced models exhibit improved robustness under adverse visual conditions such as fog, low light, or contrast reduction.

🧠 The first part of the research focuses on bio-inspired design, specifically the integration of Divisive Normalization (DN), which acts as a gain-control mechanism that regulates neuronal responses based on contextual activity.

📈 It explores how incorporating biologically inspired computations and psychophysically grounded evaluation frameworks can bridge the gap between artificial and human perception.

🔎 His research investigates perceptual alignment in artificial vision, the extent to which deep neural networks perceive and process visual information in a way comparable to humans.

📙 Dissecting the Effectiveness of Deep Features as Metric of Perceptual Image Quality
🔗

📈 It explores how incorporating biologically inspired computations and psychophysically grounded evaluation frameworks can bridge the gap between artificial and human perception.

🔎 His research investigates perceptual alignment in artificial vision, the extent to which deep neural networks perceive and process visual information in a way comparable to humans.

🔎🌏 Feini Huang is a postdoc researcher specialising in applying and evaluating Explainable AI (XAI), will use causal inference and spatio-temporal deep learning, to deepen our understanding of Earth systems.

🤖🧠 Kexuan Zhang will continue her research on machine learning, causal inference, and applications in climate analysis.

💻📊 Paolo Frazzetto will focus on developing Graph Neural Networks to address Earth System science problems, blending principles of physics-aware ML, causality, and trustworthy AI.

🌏➗ Carmen Oliver will focus her research on developing multimodal foundation models for Earth modelling.

🗺️🌧️ Paula Costa will be focused on developing risk assessment maps to better understand the impacts of climate extremes and support evidence-based decision-making.

🎉 Thanks USMILE @erc.europa.eu and XAIDA Project (xaida.eu)