Leonidas A. A. Doumas

Guillermo Puebla, Leonidas A. A. Doumas

Humans can flexibly generalize knowledge across domains by leveraging structured relational representations. While prior research has shown how such representations support analogical reasoning, less is known about how they are recruited to guide adaptive behavior. We address this gap by introducing the Relational Regression Tree Learner (RRTL), a model that incrementally builds policies over structured relational inputs by selecting task-relevant relations during the learning process. RRTL is grounded in the framework of relational reinforcement learning but diverges from traditional approaches by focusing on ground (i.e., non-variabilized) rules that refer to specific object configurations. Across three Atari games of increasing relational complexity (Breakout, Pong, Demon Attack), the model learns to act effectively by identifying a small set of relevant relations from a broad pool of candidate relations. A comparative version of the model, which partitions the state space using relative magnitude values (e.g., "more", "same", "less"), showed more robust learning than a version using logical (binary) splits. These results provide a proof of principle that reinforcement signals can guide the selection of structured representations, offering a computational framework for understanding how relational knowledge is learned and deployed in adaptive behavior.

Ivan Vegner, Sydelle de Souza, Valentin Forch et al.

A core aspect of compositionality, systematicity is a desirable property in ML models as it enables strong generalization to novel contexts. This has led to numerous studies proposing benchmarks to assess systematic generalization, as well as models and training regimes designed to enhance it. Many of these efforts are framed as addressing the challenge posed by Fodor and Pylyshyn. However, while they argue for systematicity of representations, existing benchmarks and models primarily focus on the systematicity of behaviour. We emphasize the crucial nature of this distinction. Furthermore, building on Hadley's (1994) taxonomy of systematic generalization, we analyze the extent to which behavioural systematicity is tested by key benchmarks in the literature across language and vision. Finally, we highlight ways of assessing systematicity of representations in ML models as practiced in the field of mechanistic interpretability.

Panagiotis Kourtesis, Simona Collina, Leonidas A. A. Doumas et al.

Recent research has focused on assessing either event- or time-based prospective memory (PM) using laboratory tasks. Yet, the findings pertaining to PM performance on laboratory tasks are often inconsistent with the findings on corresponding naturalistic experiments. Ecologically valid neuropsychological tasks resemble the complexity and cognitive demands of everyday tasks, offer an adequate level of experimental control, and allow a generalisation of the findings to everyday performance. The Virtual Reality Everyday Assessment Lab (VR-EAL), an immersive virtual reality neuropsychological battery with enhanced ecological validity, was implemented to comprehensively assess everyday PM (i.e., focal and non-focal event-based, and time-based). The effects of the length of delay between encoding and initiating the PM intention and the type of PM task on everyday PM performance were examined. The results revealed that everyday PM performance was affected by the length of delay rather than the type of PM task. The effect of the length of delay differentially affected performance on the focal, non-focal, and time-based tasks and was proportional to the PM cue focality (i.e., semantic relationship with the intended action). This study also highlighted methodological considerations such as the differentiation between functioning and ability, distinction of cue attributes, and the necessity of ecological validity.

Panagiotis Kourtesis, Danai Korre, Simona Collina et al.

Virtual reality (VR) head-mounted displays (HMD) appear to be effective research tools, which may address the problem of ecological validity in neuropsychological testing. However, their widespread implementation is hindered by VR induced symptoms and effects (VRISE) and the lack of skills in VR software development. This study offers guidelines for the development of VR software in cognitive neuroscience and neuropsychology, by describing and discussing the stages of the development of Virtual Reality Everyday Assessment Lab (VR-EAL), the first neuropsychological battery in immersive VR. Techniques for evaluating cognitive functions within a realistic storyline are discussed. The utility of various assets in Unity, software development kits, and other software are described so that cognitive scientists can overcome challenges pertinent to VRISE and the quality of the VR software. In addition, this pilot study attempts to evaluate VR-EAL in accordance with the necessary criteria for VR software for research purposes. The VR neuroscience questionnaire (VRNQ; Kourtesis et al., 2019b) was implemented to appraise the quality of the three versions of VR-EAL in terms of user experience, game mechanics, in-game assistance, and VRISE. Twenty-five participants aged between 20 and 45 years with 12-16 years of full-time education evaluated various versions of VR-EAL. The final version of VR-EAL achieved high scores in every sub-score of the VRNQ and exceeded its parsimonious cut-offs. It also appeared to have better in-game assistance and game mechanics, while its improved graphics substantially increased the quality of the user experience and almost eradicated VRISE. The results substantially support the feasibility of the development of effective VR research and clinical software without the presence of VRISE during a 60-minute VR session.

Panagiotis Kourtesis, Simona Collina, Leonidas A. A. Doumas et al.

Research suggests that the duration of a VR session modulates the presence and intensity of VRISE, but there are no suggestions regarding the appropriate maximum duration of VR sessions. The implementation of high-end VR HMDs in conjunction with ergonomic VR software seems to mitigate the presence of VRISE substantially. However, a brief tool does not currently exist to appraise and report both the quality of software features and VRISE intensity quantitatively. The VRNQ was developed to assess the quality of VR software in terms of user experience, game mechanics, in-game assistance, and VRISE. Forty participants aged between 28 and 43 years were recruited (18 gamers and 22 non-gamers) for the study. They participated in 3 different VR sessions until they felt weary or discomfort and subsequently filled in the VRNQ. Our results demonstrated that VRNQ is a valid tool for assessing VR software as it has good convergent, discriminant, and construct validity. The maximum duration of VR sessions should be between 55-70 minutes when the VR software meets or exceeds the parsimonious cut-offs of the VRNQ and the users are familiarized with the VR system. Also. the gaming experience does not seem to affect how long VR sessions should last. Also, while the quality of VR software substantially modulates the maximum duration of VR sessions, age and education do not. Finally, deeper immersion, better quality of graphics and sound, and more helpful in-game instructions and prompts were found to reduce VRISE intensity. The VRNQ facilitates the brief assessment and reporting of the quality of VR software features and/or the intensity of VRISE, while its minimum and parsimonious cut-offs may appraise the suitability of VR software. The findings of this study contribute to the establishment of rigorous VR methods that are crucial for the viability of immersive VR as a research and clinical tool.

Panagiotis Kourtesis, Simona Collina, Leonidas A. A. Doumas et al.

Immersive virtual reality (VR) emerges as a promising research and clinical tool. However, several studies suggest that VR induced adverse symptoms and effects (VRISE) may undermine the health and safety standards, and the reliability of the scientific results. In the current literature review, the technical reasons for the adverse symptomatology are investigated to provide suggestions and technological knowledge for the implementation of VR head-mounted display (HMD) systems in cognitive neuroscience. The technological systematic literature indicated features pertinent to display, sound, motion tracking, navigation, ergonomic interactions, user experience, and computer hardware that should be considered by the researchers. Subsequently, a meta-analysis of 44 neuroscientific or neuropsychological studies involving VR HMD systems was performed. The meta-analysis of the VR studies demonstrated that new generation HMDs induced significantly less VRISE and marginally fewer dropouts.Importantly, the commercial versions of the new generation HMDs with ergonomic interactions had zero incidents of adverse symptomatology and dropouts. HMDs equivalent to or greater than the commercial versions of contemporary HMDs accompanied with ergonomic interactions are suitable for implementation in cognitive neuroscience. In conclusion, researchers technological competency, along with meticulous methods and reports pertinent to software, hardware, and VRISE, are paramount to ensure the health and safety standards and the reliability of neuroscientific results.

Leonidas A. A. Doumas, Guillermo Puebla, Andrea E. Martin et al.

People readily generalize knowledge to novel domains and stimuli. We present a theory, instantiated in a computational model, based on the idea that cross-domain generalization in humans is a case of analogical inference over structured (i.e., symbolic) relational representations. The model is an extension of the LISA and DORA models of relational inference and learning. The resulting model learns both the content and format (i.e., structure) of relational representations from non-relational inputs without supervision, when augmented with the capacity for reinforcement learning, leverages these representations to learn individual domains, and then generalizes to new domains on the first exposure (i.e., zero-shot learning) via analogical inference. We demonstrate the capacity of the model to learn structured relational representations from a variety of simple visual stimuli, and to perform cross-domain generalization between video games (Breakout and Pong) and between several psychological tasks. We demonstrate that the model's trajectory closely mirrors the trajectory of children as they learn about relations, accounting for phenomena from the literature on the development of children's reasoning and analogy making. The model's ability to generalize between domains demonstrates the flexibility afforded by representing domains in terms of their underlying relational structure, rather than simply in terms of the statistical relations between their inputs and outputs.

Guillermo Puebla, Andrea E. Martin, Leonidas A. A. Doumas

The ability of neural networks to capture relational knowledge is a matter of long-standing controversy. Recently, some researchers in the PDP side of the debate have argued that (1) classic PDP models can handle relational structure (Rogers & McClelland, 2008, 2014) and (2) the success of deep learning approaches to text processing suggests that structured representations are unnecessary to capture the gist of human language (Rabovsky et al., 2018). In the present study we tested the Story Gestalt model (St. John, 1992), a classic PDP model of text comprehension, and a Sequence-to-Sequence with Attention model (Bahdanau et al., 2015), a contemporary deep learning architecture for text processing. Both models were trained to answer questions about stories based on the thematic roles that several concepts played on the stories. In three critical test we varied the statistical structure of new stories while keeping their relational structure constant with respect to the training data. Each model was susceptible to each statistical structure manipulation to a different degree, with their performance failing below chance at least under one manipulation. We argue that the failures of both models are due to the fact that they cannotperform dynamic binding of independent roles and fillers. Ultimately, these results cast doubts onthe suitability of traditional neural networks models for explaining phenomena based on relational reasoning, including language processing.

Andrea E. Martin, Leonidas A. A. Doumas

Humans learn complex latent structures from their environments (e.g., natural language, mathematics, music, social hierarchies). In cognitive science and cognitive neuroscience, models that infer higher-order structures from sensory or first-order representations have been proposed to account for the complexity and flexibility of human behavior. But how do the structures that these models invoke arise in neural systems in the first place? To answer this question, we explain how a system can learn latent representational structures (i.e., predicates) from experience with wholly unstructured data. During the process of predicate learning, an artificial neural network exploits the naturally occurring dynamic properties of distributed computing across neuronal assemblies in order to learn predicates, but also to combine them compositionally, two computational aspects which appear to be necessary for human behavior as per formal theories in multiple domains. We describe how predicates can be combined generatively using neural oscillations to achieve human-like extrapolation and compositionality in an artificial neural network. The ability to learn predicates from experience, to represent structures compositionally, and to extrapolate to unseen data offers an inroads to understanding and modeling the most complex human behaviors.

Leonidas A. A. Doumas, Guillermo Puebla, Andrea E. Martin

Humans readily generalize, applying prior knowledge to novel situations and stimuli. Advances in machine learning and artificial intelligence have begun to approximate and even surpass human performance, but machine systems reliably struggle to generalize information to untrained situations. We describe a neural network model that is trained to play one video game (Breakout) and demonstrates one-shot generalization to a new game (Pong). The model generalizes by learning representations that are functionally and formally symbolic from training data, without feedback, and without requiring that structured representations be specified a priori. The model uses unsupervised comparison to discover which characteristics of the input are invariant, and to learn relational predicates; it then applies these predicates to arguments in a symbolic fashion, using oscillatory regularities in network firing to dynamically bind predicates to arguments. We argue that models of human cognition must account for far-reaching and flexible generalization, and that in order to do so, models must be able to discover symbolic representations from unstructured data, a process we call predicate learning. Only then can models begin to adequately explain where human-like representations come from, why human cognition is the way it is, and why it continues to differ from machine intelligence in crucial ways.