לוח שנה משולב

Naturalistic experimental paradigms in neuroimaging arose from a pressure to test the validity of models we derive from highly controlled experiments in real-world contexts. In many cases, however, such efforts led to the realization that models developed under particular experimental manipulations failed to capture much variance outside the context of that manipulation. The critique of non-naturalistic experiments is not a recent development; it echoes a persistent and subversive thread in the history of modern psychology. The brain has evolved to guide behavior in a multidimensional world with many interacting variables. The assumption that artificially decoupling and manipulating these variables will lead to a good understanding of the brain may be untenable.
Recent advances in artificial neural networks provide an alternative computational framework to model cognition in natural contexts. In contrast to the simplified and interpretable hypotheses we test in the lab, these models do not learn simple, human-interpretable rules or representations of the world. Instead, they use local computations to interpolate over task-relevant manifolds in high-dimensional parameter space. Counterintuitively, over-parameterized deep neural models are parsimonious and straightforward, as they provide a versatile, robust solution for learning a diverse set of functions in natural contexts. Naturalistic paradigms should not be deployed as an afterthought if we hope to build models of brain and behavior that extend beyond the laboratory into the real world.
In my talk, I will discuss the relevance of deep neural models to cognition in the context of natural language and deep language models.
Zoom link-

https://weizmann.zoom.us/j/95406893197?pwd=REt5L1g3SmprMUhrK3dpUDJVeHlrZz09
Meeting ID: 954 0689 3197
Password: 750421

Hippocampal place cells fire at a high rate whenever an animal is in a specific location in an environment and are thought to support spatial and episodic memory. When an animal visits different environments, place cells typically ‘remap’ (i.e., change their preferred locations), and when revisiting the same environment, the same spatial code reemerges. In a recent study by our lab, place cells were shown to globally remap, forming multiple distinct representations (maps) of the same environment that stably coexist across time. In that study, switching between different maps of the same environment happened only after the mice were disconnected from the environment.
            Here I performed a set of experiments to further understand the mechanism underlying switching between multiple maps. My project established a way to manipulate this mechanism, both through external orientation inputs and by acting directly on the hippocampal network state using optogenetics. My results provide support for the proposed role of head-direction or other orientation signals in the switching between maps. They also support the model of maps as stable attractors, where the specific attractor (map) used depends on the initial conditions of the network.

Zoom: https://weizmann.zoom.us/j/92871200575?pwd=WWdZbXVmM1R5RkFZYnpTajloelVTZz09
Meeting ID: 928 7120 0575
Password: 344121

Many problems of interest, ranging from condensed matter physics and quantum chemistry to quantum information, require finding the ground state of a system of many interacting degrees of freedom (e.g., qubits or quantum spins). The main challenge stems from the exponential scaling of the Hilbert space dimension with the number of qubits. I will first discuss various strategies to tackle this problem using classical computers, such as tensor network states and Monte Carlo sampling, and their limitations. Quantum computers are ideally suited for this task; I will present a proposal to simulate quantum systems on noisy intermediate-scale quantum (NISQ) devices made of imperfect qubits, where the noise level translates into a finite energy density (i.e., finite temperature).

Making decisions, attending to certain items, and manipulating information in working memory are fundamental behaviors that rely on specific neural circuitry. Throughout my research I have contributed to understanding such behaviors in human and in nonhuman primates but found that despite tremendous advances in the field, we still lack a mechanistic understanding of what goes wrong in conditions such as dementia or autism. My long-term research goal is to determine the circuits that support cognitive behavior, in health and disease.
In my talk, I present three key contributions I have made towards uncovering neuronal circuits for cognition in the macaque, an animal model whose neural circuitry affords unique insight into human brain function. First, I demonstrate the utility of rigorous psychophysical frameworks in determining the causal contribution of key brain regions to behavior in a perceptual decision-making task. Next, I describe how causal manipulations of brain areas involved in attentional control can be used to identify hitherto unknown areas and reveal new functional circuits in support of selective attention and object recognition. Finally, I show how computational analyses of population data reveal circuits within circuits with distinct roles in supporting working memory.
I end the talk by presenting my future research directions and approach: to leverage my experience studying how we select from external information (from sensory signals) to investigate how we select from internal information (from information stored in visual working memory). By blending theory-driven experiments with large-scale electrophysiological recording and circuit-specific causal manipulations in behaving macaques, I aim to uncover how we select relevant information from working memory, and equally important, how we fail to do so when struck by disorders of executive or memory function.

With the growing complexity of functional materials and chemical systems, we often nd ourselves
limited in our ability to fully represent the set of descriptors of a chemical system. In complex chemical
systems, nding a complete crystallographic model that folds all the interatomic correlations using
a small set of structural descriptors may not always be feasible or practical. Alternatively, one can
take a data-driven approach and measure the relative changes in structural or chemical features (e.g,
structural correlations, oxidation states). An experimental data-driven approach does not require
complete models and enjoys the rapidly evolving machine-learning tool-set, which excel at classifying
relational datasets and, if also labelled by an observed property, can provide predictive power that
links system's descriptors with observed properties.
I will focus on two types of complexities:
(1) Hierarchical complexity, in which dierent types of structural or chemical correlations change
change with the probed correlation length. For example, in ferroic materials dierent prop-
erties (e.g., mechanical, dielectric, optoelectronic) may depend dierently on short- and long-
range structural correlations. In multi-component alloys local chemical correlations (random-
distribution, ordering, clustering) can aect corrosion and plasticity, but altogether show a single
average structural phase. Since selected materials' properties depend on correlations at a specic
hierarchical level, it is important to be able to isolate those from one another.
(2) Evolutionary complexity, where the order changes over space and/or time. Nucleation, crys-
tal growth, intercalation - are examples for processes that involve evolutionary complexity and
can also be found in batteries, heterogeneous catalysis and photovoltaics. Isolating and track-
ing order-related correlations in heterogeneous kinetically-stabilized or dynamically changing
systems is, therefore, important for their more complete understanding, design and control.
Total scattering and Pair Distribution Function (PDF) analysis are key methods for unfolding
structural correlations at dierent correlation lengths. Using 4D-STEM to generate nm-resolution
spatially-resolved electron-PDF data taken from hot-rolled Ni-laminated bulk-metallic-glass [1], I
demonstrate how both hierarchical and evolutionary complexity can be uncovered and studied. Par-
tially assisted with a machine-learning classication toolbox, we show how dierent aspects of the
structural and chemical order, such as chemical-short-range-order, can be directly visualized as a
function of position. In a dierent example [2] I show how an evolutionary complex systems can be
manipulated to achieve a desired chemical state. In this example we demonstrate an active reaction
control of Cu redox state from real-time feedback from in-situ synchrotron measurements.
While complexity can lead to a lack of control over a chemical system, it is essentially adding
tuning-knobs that, once isolated, understood and controlled, can unlock new materials with desired
functionalities.
[1] Y. Rakita, et al., Mapping Structural Heterogeneity at the Nanoscale with Scanning Nano-structure Electron Mi-
croscopy (SNEM), arXiv:2110.03589 (2021).
[2] Y. Rakita, et al., Active reaction control of Cu redox state based on real-time feedback from in situ synchrotron
measurements, JACS 142, 18758 (2020). DOI: 10.1021/jacs.0c09418.
1