Talk: The generalized drift diffusion model enables high-throughput screening of perceptual decisio… скачать в хорошем качестве

Talk: The generalized drift diffusion model enables high-throughput screening of perceptual decisio…

Speaker: Maxwell Shinn, Yale University (grid.47100.32) Title: The generalized drift diffusion model enables high-throughput screening of perceptual decision-making strategies Emcee: Doby Rahnev Backend host: Aiying Zhang Details: https://neuromatch.io/abstract?submis... Presented during Neuromatch Conference 3.0, Oct 26-30, 2020. Summary: Since the late 1970s, the drift-diffusion model (DDM) has been an important tool in the cognitive neuroscientist's toolbox, providing a way to quantify evidence accumulation during decision-making and link behavioral data to neural recordings. However, methodological challenges have limited the ability to expand the DDM to test new decision-making strategies. Specifically, the complexity of defining new models and the time required to fit them to data have prevented widespread exploration of variants on the DDM. Additionally, the DDM can only represent a limited set of experimental designs. Here, we describe our recent work on the generalized drift-diffusion model (GDDM), a framework for building and fitting DDM extensions, and a software package (PyDDM) which implements the framework (Shinn et al., 2020, eLife). Our framework makes previously intractable high-throughput approaches possible with a 100-fold or more speedup compared to standard methods. It also makes it easy to reuse different aspects of these DDM extensions, facilitated by our growing online database of model mechanisms (https://pyddm.readthedocs.io/en/lates...) to promote reproducibility and open science. Our framework allows experimentalists to fit models to more complex experimental paradigms, and theorists to conveniently formulate models and efficiently fit them with the latest methodology. We demonstrate this flexibility by analyzing a sophisticated decision-making task using a high-throughput approach to explore decision-making strategies. We presented monkeys with a perceptual decision-making task with noisy evidence extended over time that included both timing and reward manipulations. Our modeling approach discovered the key characteristics of decision-making strategies which can account for the reward and timing biases exhibited by the monkeys (Shinn et al., 2020, J Neurosci). Overall, we show that the GDDM in conjunction with PyDDM enables new possibilities for understanding decision-making.