Periodic Reporting for period 3 - PIVOTAL (Predictive Memory Systems Across the Human Lifespan) Reporting period: 2021-02-01 to 2022-07-31 Summary of the context and overall objectives of the project Imagine unlocking your office in the morning. Within milliseconds you will be able to tell whether the furniture, computer, and papers on the desk are still where they are expected to be. You will also quickly detect if something unexpected is in the room, for example a box of chocolates an appreciative colleague has left for you. The mechanism proposed to underlie our mind’s efficient ability to grasp the environment converges on the notion of predictive brain. Put forward as a universal principle of the brain, the crux of the notion is that brains are essentially prediction machines that constantly attempt to match incoming inputs with top-down predictions. This provides us a powerful means to streamline the massive amount of continuous incoming information from the environment. Furthermore, when actual input is discrepant from the predicted input, a prediction error (PE) is elaborated to drive learning, i.e. updating internal models that will help to improve future predictions.If the predictive brain is indeed a unifying principle, two critical issues need to be resolved. First, the predictive coding framework has not yet delineated the nature of internal models on which predictions are based (e.g. memory of prior experience) and how our actual experiences shape them in turn. Second, how does such a universal brain principle play out in diverse brains (e.g. young versus old brains)? Addressing these knowledge gaps is important in order to make a breakthrough in our understanding of the fundamental nature of the human mind and brain and to test the adaptivity of the predictive brain principle in accommodating inherent diversities of human brains.By connecting three separate strands of research (i.e. predictive coding, memory systems, and lifespan development), the PIVOTAL research program aims to unravel the cognitive and neural mechanisms that enable the brain to (i) generate predictions based on memory of prior experience (episodic memory) and knowledge about the world (semantic memory); (ii) verify its predictions given the actual event, and (iii) engage in subsequent processes that in turn modify the memory representation. Using cognitive neuroscience methodology (functional magnetic resonance imaging) and experimental research designs, these mechanisms are being systematically examined in children, younger adults, and older adults, whose neurocognitive landscapes are highly different from each other. The gain in knowledge will characterize the cognitive architectures that allow the human brain to perform predictive processing as a fundamental operation in its interaction with the environment. Work performed from the beginning of the project to the end of the period covered by the report and main results achieved so far The current report covers the first 30 month period of the project that lasts for 5 years. Therefore, it reflects our interim progress towards the overall goals of the project. In relation to prediction generation, we started an fMRI study (3 Tesla scanner), in which young adult participants are being shown partially occluded visual scenes varying by episodic or semantic content. Episodic scenes contain unrelated objects, while semantic scenes contain combinations of objects that cue for a specific schema (e.g. blackboard, tables, chairs —classroom). The cortical representation of the target occluded region in visual cortex areas (V1 and 2) is being mapped in each individual, allowing us to isolate feedback/prediction signals in these area without any incoming signal from the environment. For the analysis, pattern classifiers from a machine-learning algorithm are being trained to learn the mapping between the multivariate brain-activity patterns of the target brain areas and the presented scenes. The classifiers are then being tested on an independent set of test data to decode which scene was presented, with the classification accuracy providing a measure of contextual feedback/prediction signal. Our preliminary results show that the classifiers achieve above chance accuracy in distinguishing objects that are being occluded in the scene, both when the source of information come from episodic memory and semantic memory. As the next step, we are running a study with similar design at a 7 Tesla scanner, which allows us to attain cutting-edge, high-resolution scans for delineation of the hippocampus as well as layers within the V1 areas. With this, we will pinpoint the specific layers within V1 at which prediction signals from the hippocampus (presumably the CA3/CA1 due to their roles in pattern completion) arrive.In relation to prediction verification and subsequent memory processing, the first step we took here is to develop a paradigm that allows us to reliably measure the relationship between predictive processing and memory performance. In our paradigm, we trained young adults in associations between certain objects and contexts (e.g. musical instruments are usually found on the beach). Once these associations are acquired, we ask our participants to use them to make predictions about objects in a given context and systematically match or mismatch their expectations. Later on, we use a recognition memory test to assess the quality of the memory traces generated during those events that vary in prediction error. We ran several pilot studies and the results generally support the differential impact of different degrees of PE in shaping new memories and highlight the importance of memory consolidation in the process.In addition to more classical approaches on behavioral data analysis, we also started using computational models that allow the quantification of PE and prior expectations. The estimations of these quantities, trial-by-trial, will permit to relate PE and priors to memory formation, which in turn will stimulate the search for their neural correlates. Preliminary results from this computational work suggests that episodic remembering is indeed related to PE. Precisely, when outcomes were better than expected (positive PE) participants’ ability to remember the items in the subsequent recognition test was improved, compared to items presented in conditions where outcomes were worse than expected (negative PE). Progress beyond the state of the art and expected potential impact (including the socio-economic impact and the wider societal implications of the project so far) Understanding how the human brain operates as a prediction machine continues to be a hot topic. However, to our knowledge we remain the the only group systematically examining the mechanisms that underlie predictive processing in relation to episodic memory and semantic memory, and doing so within a lifespan context. Our findings will reveal the nature of internal models in the predictive coding framework with an explicit established link to human memory systems. By doing this we expand the horizons of both the predictive coding and memory fields. We will also have a better understanding of how these mechanisms operate in children, younger adults, and older adults, who differ from each other in important ways due to divergence in developmental orientation (progression vs. conservation) and neurocognitive landscape (structural and functional integrity of memory neural circuits). We will contribute a more dynamic perspective to the predictive brain theory that can be applied to the human lifespan and beyond. This can advance the potential to understand developmental disorders and pathological aging where predictive processing become aberrant, addressing why certain disorders tend to emerge in particular time windows (e.g. schizophrenia during adolescence).
