Gromov-Wasserstein unsupervised alignment reveals structural correspondences between the color similarity structures of humans and large language models.

Large Language Models (LLMs), such as the General Pre-trained Transformer (GPT), have shown remarkable performance in various cognitive tasks. However, it remains unclear whether these models have the ability to accurately infer human perceptual representations. Previous research has addressed this question by quantifying correlations between similarity response patterns of humans and LLMs. Correlation provides a measure of similarity, but it relies pre-defined item labels and does not distinguish category- and item- level similarity, falling short of characterizing detailed structural correspondence between humans and LLMs. To assess their structural equivalence in more detail, we propose the use of an unsupervised alignment method based on Gromov-Wasserstein optimal transport (GWOT). GWOT allows for the comparison of similarity structures without relying on pre-defined label correspondences and can reveal fine-grained structural similarities and differences that may not be detected by simple correlation analysis. Using a large dataset of similarity judgments of 93 colors, we compared the color similarity structures of humans (color-neurotypical and color-atypical participants) and two GPT models (GPT-3.5 and GPT-4). Our results show that the similarity structure of color-neurotypical participants can be remarkably well aligned with that of GPT-4 and, to a lesser extent, to that of GPT-3.5. These results contribute to the methodological advancements of comparing LLMs with human perception, and highlight the potential of unsupervised alignment methods to reveal detailed structural correspondences.

Are Color Experiences the Same across the Visual Field?

It seems obvious to laypeople that neurotypical humans experience color equivalently across their entire visual field. To some neuroscientists, psychologists, and philosophers, though, this claim has been met with skepticism, as neurophysiological evidence indicates the mechanisms that support color perception degrade with eccentricity. However, the argument that this entails altered color experience in peripheral vision is not universally accepted. Here, we address whether color experience is essentially equivalent between central and peripheral vision. To assess this, we will obtain similarity relationships between color experiences across the visual field using both online and laboratory-based far-field displays, while removing the confounds of saccades, memory, and expectation about color experiences. Our experiment was designed to provide clear evidence that would favor either unchanged or altered color experience relationships in the periphery. Our results are consistent with lay people's phenomenological reports: Color experiences, as probed by similarity relationships in central vision and the far field (60°), are equivalent when elicited by large stimuli. These findings challenge the widespread view in philosophy and cognitive science that peripheral color experiences are illusory, and are discussed in the context of their related neurophysiological, psychophysical, and philosophical literature.

A Quantum Geometric Framework for Modeling Color Similarity Judgments.

Since Tversky argued that similarity judgments violate the three metric axioms, asymmetrical similarity judgments have been particularly challenging for standard, geometric models of similarity, such as multidimensional scaling. According to Tversky, asymmetrical similarity judgments are driven by differences in salience or extent of knowledge. However, the notion of salience has been difficult to operationalize, especially for perceptual stimuli for which there are no apparent differences in extent of knowledge. To investigate similarity judgments between perceptual stimuli, across three experiments, we collected data where individuals would rate the similarity of a pair of temporally separated color patches. We identified several violations of symmetry in the empirical results, which the conventional multidimensional scaling model cannot readily capture. Pothos et al. proposed a quantum geometric model of similarity to account for Tversky's findings. In the present work, we extended this model to a more general framework that can be fit to similarity judgments. We fitted several variants of quantum and multidimensional scaling models to the behavioral data and concluded in favor of the quantum approach. Without further modifications of the model, the best-fit quantum model additionally predicted violations of the triangle inequality that we observed in the same data. Overall, by offering a different form of geometric representation, the quantum geometric framework of similarity provides a viable alternative to multidimensional scaling for modeling similarity judgments, while still allowing a convenient, spatial illustration of similarity.

A Preclinical Model of Computerized Cognitive Training: Touchscreen Cognitive Testing Enhances Cognition and Hippocampal Cellular Plasticity in Wildtype and Alzheimer's Disease Mice.

With the growing popularity of touchscreen cognitive testing in rodents, it is imperative to understand the fundamental effects exposure to this paradigm can have on the animals involved. In this study, we set out to assess hippocampal-dependant learning in the APP/PS1 mouse model of Alzheimer's disease (AD) on two highly translatable touchscreen tasks - the Paired Associate Learning (PAL) task and the Trial Unique Non-Matching to Location (TUNL) task. Both of these tests are based on human tasks from the Cambridge Neuropsychological Test Automated Battery (CANTAB) and are sensitive to deficits in both mild cognitive impairment (MCI) and AD. Mice were assessed for deficits in PAL at 9-12 months of age, then on TUNL at 8-11 and 13-16 months. No cognitive deficits were evident in APP/PS1 mice at any age, contrary to previous reports using maze-based learning and memory tasks. We hypothesized that daily and long-term touchscreen training may have inadvertently acted as a cognitive enhancer. When touchscreen-tested mice were assessed on the Morris water maze, they showed improved task acquisition compared to naïve APP/PS1 mice and wild-type (WT) littermate controls. In addition, we show that touchscreen-trained WT and APP/PS1 mice show increased cell proliferation and immature neuron numbers in the dentate gyrus compared to behaviorally naïve WT and APP/PS1 mice. This result indicates that the touchscreen testing paradigm could improve cognitive performance, and/or mask an impairment, in experimental mouse models. This touchscreen-induced cognitive enhancement may involve increased neurogenesis, and possibly other forms of cellular plasticity. This is the first study to show increased numbers of proliferating cells and immature neurons in the hippocampus following touchscreen testing, and that touchscreen training can improve cognitive performance in maze-based spatial navigation tasks. This potential for touchscreen testing to induce cognitive enhancement, or other phenotypic shifts, in preclinical models should be considered in study design. Furthermore, touchscreen-mediated cognitive enhancement could have therapeutic implications for cognitive disorders.

Progressive impairments in executive function in the APP/PS1 model of Alzheimer's disease as measured by translatable touchscreen testing.

Executive function deficits in Alzheimer's disease (AD) occur early in disease progression and may be predictive of cognitive decline. However, no preclinical studies have identified deficits in rewarded executive function in the commonly used APPSwe/PS1∆E9 (APP/PS1) mouse model. To address this, we assessed 12-26 month old APP/PS1 mice on rewarded reversal and/or extinction tasks. 16-month-old, but not 13- or 26-month-old, APP/PS1 mice showed an attenuated rate of extinction. Reversal deficits were seen in 22-month-old, but not 13-month-old APP/PS1 animals. We then confirmed that impairments in reversal were unrelated to previously reported visual impairments in both AD mouse models and humans. Age, but not genotype, had a significant effect on markers of retinal health, indicating the deficits seen in APP/PS1 mice were directly related to cognition. This is the first characterisation of rewarded executive function in APP/PS1 mice, and has great potential to facilitate translation from preclinical models to the clinic.

Microbiome Profiling Reveals Gut Dysbiosis in the Metabotropic Glutamate Receptor 5 Knockout Mouse Model of Schizophrenia.

Schizophrenia (SZ) is a psychiatric disorder that constitutes one of the top 10 global causes of disability. More recently, a potential pathogenic role for the gut microbial community (microbiota) has been highlighted, with numerous studies describing dysregulated microbial profiles in SZ patients when compared to healthy controls. However, no animal model of SZ has previously recapitulated the gut dysbiosis observed clinically. Since the metabotropic glutamate receptor 5 (mGlu5) knockout mice provide a preclinical model of SZ with strong face and predictive validity, in the present study we performed gut microbiome profiling of mGlu5 knockout (KO) and wild-type (WT) mice by 16S rRNA sequencing of bacterial genomic DNA from fecal samples, analyzing bacterial diversity and taxonomic composition, as well as gastrointestinal parameters as indicators of gut function. We found a significant genotype difference in microbial beta diversity. Analysis of composition of microbiomes (ANCOM) models were performed to evaluate microbiota compositions, which identified a decreased relative abundance of the Erysipelotrichaceae family and Allobaculum genus in this mouse model of SZ. We also identified a signature of bacteria discriminating between the genotypes (KO and WT), consisting of the Erysipelotrichales, Bacteroidales, and Clostridiales orders and macroscopic gut differences. We thus uncovered global differential community composition in the gut microbiota profile between mGlu5 KO and WT mice, outlining the first evidence for gut dysbiosis in a genetic animal model of SZ. Our findings suggest that this widely used preclinical model of SZ also has substantial utility for investigations of gut dysbiosis and associated signaling via the microbiota-gut-brain axis, as potential modulators of SZ pathogenesis. Our discovery opens up new avenues to explore gut dysbiosis and its proposed links to brain dysfunction in SZ, as well as novel therapeutic approaches to this devastating disorder.

Paradoxical effects of exercise on hippocampal plasticity and cognition in mice with a heterozygous null mutation in the serotonin transporter gene.

BACKGROUND AND PURPOSE: Exercise is known to improve cognitive function, but the exact synaptic and cellular mechanisms remain unclear. We investigated the potential role of the serotonin (5-HT) transporter (SERT) in mediating these effects. EXPERIMENTAL APPROACH: Hippocampal long-term potentiation (LTP) and neurogenesis were measured in standard-housed and exercising (wheel running) wild-type (WT) and SERT heterozygous (HET) mice. We also assessed hippocampal-dependent cognition using the Morris water maze (MWM) and a spatial pattern separation touchscreen task. KEY RESULTS: SERT HET mice had impaired hippocampal LTP regardless of the housing conditions. Exercise increased hippocampal neurogenesis in WT mice. However, this was not observed in SERT HET animals, even though both genotypes used the running wheels to a similar extent. We also found that standard-housed SERT HET mice displayed altered cognitive flexibility than WT littermate controls in the MWM reversal learning task. However, SERT HET mice no longer exhibited this phenotype after exercise. Cognitive changes, specific to SERT HET mice in the exercise condition, were also revealed on the touchscreen spatial pattern separation task, especially when the cognitive pattern separation load was at its highest. CONCLUSIONS AND IMPLICATIONS: Our study is the first evidence of reduced hippocampal LTP in SERT HET mice. We also show that functional SERT is required for exercise-induced increase in adult neurogenesis. Paradoxically, exercise had a negative impact on hippocampal-dependent cognitive tasks, especially in SERT HET mice. Taken together, our results suggest unique complex interactions between exercise and altered 5-HT homeostasis.

Transgenic Mouse Models as Tools for Understanding How Increased Cognitive and Physical Stimulation Can Improve Cognition in Alzheimer's Disease.

Cognitive decline appears as a core feature of dementia, of which the most prevalent form, Alzheimer's disease (AD) affects more than 45 million people worldwide. There is no cure, and therapeutic options remain limited. A number of modifiable lifestyle factors have been identified that contribute to cognitive decline in dementia. Sedentary lifestyle has emerged as a major modifier and accordingly, boosting mental and physical activity may represent a method to prevent decline in dementia. Beneficial effects of increased physical activity on cognition have been reported in healthy adults, showing potential to harness exercise and cognitive stimulation as a therapy in dementia. 'Brain training' (cognitive stimulation) has also been investigated as an intervention protecting against cognitive decline with normal aging. Consequently, the utility of exercise regimes and/or cognitive stimulation to improve cognition in dementia in clinical populations has been a major area of study. However, these therapies are in their infancy and efficacy is unclear. Investigations utilising animal models, where dose and timing of treatment can be tightly controlled, have provided many mechanistic insights. Genetically engineered mouse models are powerful tools to investigate mechanisms underlying cognitive decline, and also how environmental manipulations can alter both cognitive outcomes and pathology. A myriad of effects following physical activity and housing in enriched environments have been reported in transgenic mice expressing Alzheimer's disease-associated mutations. In this review, we comprehensively evaluate all studies applying environmental enrichment and/or increased physical exercise to transgenic mouse models of Alzheimer's disease. It is unclear whether interventions must be applied before first onset of cognitive deficits to be effective. In order to determine the importance of timing of interventions, we specifically scrutinised studies exposing transgenic mice to exercise and environmental enrichment before and after first report of cognitive impairment. We discuss the strengths and weaknesses of these preclinical studies and suggest approaches for enhancing rigor and using mechanistic insights to inform future therapeutic interventions.

Touchscreen testing reveals clinically relevant cognitive abnormalities in a mouse model of schizophrenia lacking metabotropic glutamate receptor 5.

Metabotropic glutamate receptor 5 (mGlu5) has been implicated in certain forms of synaptic plasticity and cognitive function. mGlu5 knockout (KO) mice and mGlu5 antagonists have been previously used to study the pathophysiology of schizophrenia as they have been shown respectively to display or induce endophenotypes relevant to schizophrenia. While schizophrenia presents with generalized cognitive impairments, the cognitive phenotype of mice lacking mGlu5 has so far only been explored using largely hippocampal-dependent spatial and contextual memory tasks. To address this, we used a touchscreen system to assess mGlu5 KO mice for pairwise visual discrimination, reversal learning, and extinction of an instrumental response requiring no discrimination. Furthermore, we tested the role of mGlu5 in working memory using the Trial-Unique Non-Matching to Location (TUNL) task utilizing pharmacological ablation. mGlu5 KO mice were impaired on discrimination learning, taking longer to reach criterion and requiring more correction learning trials. Performance on reversal learning was also impaired, with mGlu5 KO mice demonstrating a perseverative phenotype. The mGlu5 KO mice responded at a higher rate during extinction, consistent with this perseverative profile. In contrast, wildtype mice treated acutely with an mGlu5 antagonist (MTEP) showed no deficits in a touchscreen task assessing working memory. The present study demonstrates learning and memory deficits as well as an increased perseverative phenotype following constitutive loss of mGlu5 in this mouse model of schizophrenia. These findings will inform translational approaches using this preclinical model and the pursuit of mGlu5 as therapeutic target for schizophrenia and other brain disorders.

Environmental enrichment enhances cognitive flexibility in C57BL/6 mice on a touchscreen reversal learning task.

Environmental enrichment (EE) is any positive modification of the 'standard housing' (SH) conditions in which laboratory animals are typically held, usually involving increased opportunity for cognitive stimulation and physical activity. EE has been reported to enhance baseline performance of wild-type animals on traditional cognitive behavioural tasks. Recently, touchscreen operant testing chambers have emerged as a way of performing rodent cognitive assays, providing greater reproducibility, translatability and automatability. Cognitive tests in touchscreen chambers are performed over numerous trials and thus experimenters have the power to detect subtle enhancements in performance. We used touchscreens to analyse the effects of EE on reversal learning, visual discrimination and hippocampal-dependent spatial pattern separation and working memory. We hypothesized that EE would enhance the performance of mice on cognitive touchscreen tasks. Our hypothesis was partially supported in that EE induced enhancements in cognitive flexibility as observed in visual discrimination and reversal learning improvements. However, no other significant effects of EE on cognitive performance were observed. EE decreased the activity level of mice in the touchscreen chambers, which may influence the enrichment level of the animals. Although we did not see enhancements on all hypothesized parameters, our testing paradigm is capable of detecting EE-induced improved cognitive flexibility in mice, which has implications for both understanding the mechanisms of EE and improving screening of putative cognitive-enhancing therapeutics.