Speakers
Workshop Speakers
The workshop will consist of talks (25 min + 5 min questions) given by the following list of invited speakers (in alphabetical order). The research presented during the workshop will span across species (humans, rodents), but also across disciplines (psychology, neuroscience, economics) and methodologies (electrophysiology, neuroimaging, pharmacology, genetics, modeling).
Rahul Bhui
Harvard University, USA
Context dependence and efficient neural coding with noisy samples
It is well known that the way we feel about an outcome depends on how we compare it to our past and present experiences. However, it is not so clear what kind of comparisons we make, or even why we make comparisons at all. In this talk, I draw on the neurocomputational principle of efficient coding to explain why this context dependence is adaptive, and to show how the optimal coding scheme changes in the presence of noise. I discuss how influential psychological theories of context-sensitive judgment can be derived from the efficient coding framework. The amount of information that can be transmitted by neurons is maximized by the rank transformation of "decision by sampling", while lost efficiency due to noise in the retrieval of experiences can be counteracted by the range component of "range-frequency theory". This provides a unified neurally grounded basis for some of the most seminal theories of judgment and decision making.
Nicola Grissom
University of Minnesota, USA
Genetic contributions to variable strategies in reinforcement learning
Autism spectrum disorders comprise a diverse set of behaviors, including restricted interests, repetitive behaviors, and challenges in complex social environments. We now recognize that the core features of autism can be united as reflecting challenges with prediction - correctly anticipating the most likely outcomes of our behavior. Accordingly, people on the autism spectrum select very different strategic approaches than neurotypical people while performing reinforcement learning and decision making tasks, and these strategic approaches, depending on the task design, can produce advantages or disadvantages for people on the spectrum compared to neurotypical people. How and why autism alters these strategic approaches is our research focus. Autism is highly genetic. In my lab, we have established that male mice carrying autism-associated genotypes also show a different strategic approach to reinforcement learning compared to control males, but that female carriers of these genotypes are largely similar to control females. This is consistent with the fact that women and girls don't seem to have the same experience of autism as men and boys do, and are diagnosed less frequently than men and boys. We find that male mice carrying autism-associated genotypes have a greater disruption of molecular mechanisms in brain regions supporting reward prediction, indicating female protection at a molecular level. However, in bandit tasks, we find that female and male control mice produce very different behavioral strategies that lead to faster acquisition in females, suggesting that autism associated genotypes interact with sex differences in the neural mechanisms of reward-guided decision making to produce male vulnerability.
Falk Lieder
Max Planck Institute for Intelligent Systems, Tübingen, Germany
Resource-rational decision-making
A substantial literature in psychology and behavioral economics has demonstrated that decision makers systematically violate the prescriptions of expected utility theory. Despite considerable efforts, our understanding of the underlying cognitive mechanisms is still fuzzy and disjointed. In this talk, I present a theoretical framework that integrates the psychological realism of heuristics and biases with the unifying power and mathematical precision of normative principles. The essence of this theory is that decision-makers should make optimal use of their finite time and limited cognitive resources (resource-rationality). This theory leads to an automatic method for deriving rational heuristics from first principles. I will illustrate that combining this method with process tracing allows us to answer descriptive, normative, and prescriptive questions about people’s decision strategies. In the first part of my talk, I illustrate that our method can be used to discover and make sense of people’s decision strategies and their variability. In the second part of my talk, I will introduce a method for measuring the learning-induced variability in people's planning strategies and compare the measured changes in people's planning strategies to rational models of strategy discovery.
Rafael Polania
ETH Zürich, Switzerland
Efficient noisy sampling and decision behaviour
Sampling is ubiquitous in neural circuits starting from ion-channel gating and generation of action potentials, all the way to behavioral decisions. However, it remains unclear why samples of information guiding decisions are remarkably noisy, thus causing variable and apparent irrational behavior. Moreover, how the nervous system simultaneously considers regularities in the environment, goals of the organism, and capacity constraints to guide behavior via information sampling remains unknown. Here, we formally demonstrate that optimal decision behavior in capacity-limited systems can be parsimoniously achieved via discrete samples, where crucially, noise serves to optimize information transfer of the environmental regularities while accounting for the specific goals of the individuals (e.g., do I care about representing the world accurately or do I care about maximizing reward/fitness). We apply this theory to test the long-held hypothesis that, due to evolutionary processes, humans may efficiently adapt their non-symbolic numerosity system to maximize fitness when making decisions. Surprisingly, we found that humans do not directly follow these resource-limited optimal recipes (maximize neither accuracy nor fitness), but appear to rely on "suboptimal" but efficient proxy mechanisms of sampling from memory irrespective of the organism’s goal. These findings provide a general mechanistic framework for understanding decision behavior while accounting for biological restrictions of information coding.
Konstantinos Tsetsos
University Medical Center Hamburg-Eppendorf, Germany
Overcoming decision noise via selective evidence integration
Years of research in psychology and neuroscience have shown that when making decisions, humans and other animals engage in serial sampling and gradual accumulation of evidence. From a decision theoretic perspective and in stationary environments, this process of gradual accumulation can maximise choice accuracy. Evidence accumulation is thus widely considered to be the key mechanism via which biological brains approximate decision optimality. Here, I will show that in biological brains, where noise can act at different processing stages, this mechanism is necessary but not sufficient to achieving decision optimality. Instead, a more nuanced evidence accumulation algorithm — that samples evidence for different alternatives concurrently and discounts momentarily weaker samples prior to accumulation — outperforms perfect evidence accumulation. The advantage of this "selective integration" mechanism is maximally realised when the extent of evidence discounting is proportional to the level of "decision" noise, i.e. the noise that occurs downstream sensory representation. Importantly, selective integration can lead to decisions that are at odds with rational choice theory, namely with the principles of transitivity and regularity. I will present data showing that indeed humans violate these rationality principles but, crucially, they do so in proportion to the levels of their decision noise. Taken together, these results offer a normative solution to a century-long conundrum in decision sciences, showing that decision irrationalities are by-products of adaptive computational mechanisms.
Anne Urai
Cold Spring Harbor Laboratory, USA
Choice history biases subsequent evidence accumulation
Perceptual choices not only depend on the current sensory input, but also on the behavioral context, such as the history of one’s own choices. Yet, it remains unknown how such history signals shape the dynamics of later decision formation. In models of decision formation, it is commonly assumed that choice history shifts the starting point of accumulation towards the bound reflecting the previous choice. I will present results that challenge this idea. By fitting bounded-accumulation decision models to behavioral data from perceptual choice tasks, we estimated bias parameters that depended on observers’ previous choices. Across multiple task protocols and sensory modalities, individual history biases in overt behavior were consistently explained by a history-dependent change in the evidence accumulation, rather than in its starting point. Choice history signals thus seem to bias the interpretation of current sensory input, akin to shifting endogenous attention towards (or away from) the previously selected interpretation.
Valentin Wyart
Inserm – Ecole Normale Supérieure, Paris, France
Curious by choice or by chance? Learning noise drives decision variability in volatile environments
When learning the value of actions in volatile environments, humans make a sizable fraction of 'non-greedy' decisions which do not maximize expected value. Prominent theories describe these decisions as the result of a compromise between exploiting a currently well-valued action vs. exploring more uncertain alternatives. However, we have recently shown that the variability of perceptual decisions based on multiple cues is bounded not by sensory errors nor by choice stochasticity, but by computational noise in probabilistic inference. We thus reasoned that a substantial fraction of non-greedy decisions may be caused by the same kind of internal noise. We derived a formulation of reinforcement learning which allows for internal noise in its core computations. And in a series of behavioral, neuroimaging, pupillometric and pharmacological experiments, we quantified the impact of learning noise on behavior and identified its neural substrates. At the behavioral level, we show that more than half of non-greedy decisions are triggered by learning noise. At the neural level, the amount of learning noise can be predicted by BOLD activity in the anterior cingulate cortex and by pupillary dilation, and increased by pharmacological manipulation of cortical norepinephrine levels. These findings indicate that the decision variability observed in volatile environments is driven primarily not by exploratory choices, but by the limited precision of reward-guided learning.