Christian F. Doeller

Can Demircan, Tankred Saanum, Leonardo Pettini et al.

Humans represent scenes and objects in rich feature spaces, carrying information that allows us to generalise about category memberships and abstract functions with few examples. What determines whether a neural network model generalises like a human? We tested how well the representations of $86$ pretrained neural network models mapped to human learning trajectories across two tasks where humans had to learn continuous relationships and categories of natural images. In these tasks, both human participants and neural networks successfully identified the relevant stimulus features within a few trials, demonstrating effective generalisation. We found that while training dataset size was a core determinant of alignment with human choices, contrastive training with multi-modal data (text and imagery) was a common feature of currently publicly available models that predicted human generalisation. Intrinsic dimensionality of representations had different effects on alignment for different model types. Lastly, we tested three sets of human-aligned representations and found no consistent improvements in predictive accuracy compared to the baselines. In conclusion, pretrained neural networks can serve to extract representations for cognitive models, as they appear to capture some fundamental aspects of cognition that are transferable across tasks. Both our paradigms and modelling approach offer a novel way to quantify alignment between neural networks and humans and extend cognitive science into more naturalistic domains.

Markus Frey, Christian F. Doeller, Caswell Barry

Deep artificial neural networks (DNNs) trained through backpropagation provide effective models of the mammalian visual system, accurately capturing the hierarchy of neural responses through primary visual cortex to inferior temporal cortex (IT). However, the ability of these networks to explain representations in higher cortical areas is relatively lacking and considerably less well researched. For example, DNNs have been less successful as a model of the egocentric to allocentric transformation embodied by circuits in retrosplenial and posterior parietal cortex. We describe a novel scene perception benchmark inspired by a hippocampal dependent task, designed to probe the ability of DNNs to transform scenes viewed from different egocentric perspectives. Using a network architecture inspired by the connectivity between temporal lobe structures and the hippocampus, we demonstrate that DNNs trained using a triplet loss can learn this task. Moreover, by enforcing a factorized latent space, we can split information propagation into "what" and "where" pathways, which we use to reconstruct the input. This allows us to beat the state-of-the-art for unsupervised object segmentation on the CATER and MOVi-A,B,C benchmarks.

Leonardo Pettini, Carsten Bogler, Christian Doeller et al.

Naturalistic scenes are of key interest for visual perception, but controlling their perceptual and semantic properties is challenging. Previous work on naturalistic scenes has frequently focused on collections of discrete images with considerable physical differences between stimuli. However, it is often desirable to assess representations of naturalistic images that vary along a continuum. Traditionally, perceptually continuous variations of naturalistic stimuli have been obtained by morphing a source image into a target image. This produces transitions driven mainly by low-level physical features and can result in semantically ambiguous outcomes. More recently, generative adversarial networks (GANs) have been used to generate continuous perceptual variations within a stimulus category. Here we extend and generalize this approach using a different machine learning approach, a text-to-image diffusion model (Stable Diffusion XL), to generate a freely customizable stimulus set of photorealistic images that are characterized by gradual transitions, with each image representing a unique exemplar within a prompted category. We demonstrate the approach by generating a set of 108 object scenes from 6 categories. For each object scene, we generate 10 variants that are ordered along a perceptual continuum. This ordering was first estimated using a machine learning model of perceptual similarity (LPIPS) and then subsequently validated with a large online sample of human participants. In a subsequent experiment we show that this ordering is also predictive of confusability of stimuli in a working memory experiment. Our image set is suited for studies investigating the graded encoding of naturalistic stimuli in visual perception, attention, and memory.