לוח שנה משולב

In a couple of years from now, we will finish kernel programming for Spinach – a spin dynamics simulation library that supports all types of magnetic resonance spectroscopy, from Gd3+ DEER, through DNP and NMR, and all the way to singlet state diffusion MRI, including chemical kinetics, optimal control, and advanced relaxation theories. This level of generality hinges on:

1. The ability to treat classical degrees of freedom (diffusion, hydrodynamics, radiofrequency and microwave phases, stochastic tumbling, etc.) at the same conceptual level as spin degrees of freedom – the corresponding classical equations of motion must be integrated into the density matrix formalism.

2. The ability to survive enormous Kronecker products. A well digitised medical phantom would have at least a hundred points in each of the three directions, meaning a dimension of at least 1003 = 106 for the spatial dynamics generator matrices. At the same time, a typical radical contains upwards of ten coupled spins, meaning a Liouville space dimension of at least 410. Direct products of spin and spatial dynamics generators would then have the dimension in excess of 1012 even before chemical kinetics is considered.

3. Code parallelisation over cluster architectures, including the possibility of using a GPU on each node of the cluster. The principal problem is parallelisation mode switching between powder averages, indirect dimensions of pulse sequences, frequency points of frequency domain simulations, etc. – each simulation type would in general require a different mode of parallelisation and GPU utilisation.

This report is about solving all of this, and on where the dark art of simulating a time-domain magnetic resonance experiment stands at the moment. Two recent innovations are the abandonment of Liouville equation in favour of Fokker-Planck equation as the core formalism, and the use of tensor structured objects that never open Kronecker products. A separate story is recent GPUs: NVidia Tesla V100 performs ~1013 double-precision multiplications per second – an astounding amount of computing power that is surprisingly easy to use.

The preservation of the organic matter (OM) occurs as a result of post-depositional abiotic sulfurization, condensation and polymerization processes that convert the OM into stable macromolecular material termed kerogen. Different sulfurization processes, pathways and rates affect the 34S values of organic and inorganic S compounds. These sulfurization processes are affected by the redox conditions and paleo-environmental conditions (e.g. organic matter and Fe availability). Therefore, studying the organic and inorganic S distribution and their associated 34S values could be useful for understanding the paleo-environmental history associated with the deposition of ancient organic rich sediments. Until recently, only bulk phases of S could be measured for their 34S values, usually excluding organic S. A new method was developed that allows for S isotope analysis of specific organic S compounds (OSCs) at the sub- nanogram level. In my talk I will give an overview about the utility of compound specific S isotope analysis (CSSIA) for the study of different geochemical environments (e.g., immature organic rich sediments). Applying CSSIA to immature organic rich sediments from the Monterey and Ghareb formations I will show the combination of biomarkers and their S isotope composition in a single analysis. This provides a more detailed and in-depth understanding of the S and C cycles than bulk measurements of organic and inorganic S species alone.

The macroscopic fluctuation theory gives an efficient hydrodynamic description for classical nonequilibrium diffusive systems. In this talk, we would cover how it can be applied and generalised in three directions:
a. Towards a theory for open quantum diffusive systems, comparable to the macroscopic fluctuation theory.
b. Defect formation as a system is (slowly) driven in time through a continuous phase transition can be described by a scaling theory - the Kibble-Zurek Mechanism. The macroscopic fluctuation theory allows to explore the exact evolution of defects for a large set of cases. Thus, we are in a position to go beyond the scaling arguments of the Kibble-Zurek Mechanism.
c. The recently discovered thermodynamic uncertainty relations define a transport efficiency in thermal systems showing that the mean current, current fluctuations and dissipation are intimately linked. Here we will briefly show how this idea can be extended to (athermal) mesoscopic coherent processes.

Face recognition is a computationally challenging classification task that requires generalization across different views of the same identity as well as discrimination across different identities of a relatively homogenous set of visual stimuli. How does the brain resolve this taxing classification task? It is well-established that faces are processed by specialized neural mechanisms in high-level visual cortex. Nevertheless, it is not clear how this divergence to a face-specific and an object-general system contributes to face recognition. Recent advances in machine face recognition together with our understanding of how humans recognize faces enable us to address this question. In particular, I will show that a deep convolutional neural network (DCNN) that is trained on face recognition, but not a DCNN that is trained on object recognition, is sensitive to the same view-invariant facial features that humans use for face recognition. Similar to the hierarchical architecture of the visual system that diverges to a face and an object system at high-level visual cortex, a human-like, view-invariant face representation emerges only at higher layers of the face-trained but not the object-trained neural network. This view-invariant face representation is specific to the category of faces that the system was trained with both in humans and machines. I will therefore further emphasize the important role of experience and suggest that human face recognition depends on our social experience with familiar faces (“supervised learning”) rather than passive perceptual exposure to unfamiliar faces (“unsupervised learning”), highlighting the important role of social cognition in face recognition.

Additive manufacturing, which is fabrication through printing processes, has gained a lot of interest in the academy and industry, and is considered as the next industrial revolution. The synthesis and formulations of new inks compositions will be presented, along with their applications in various fields. New materials and processes for 2, 3, and 4D printing will be introduced, for fabrication of objects composed of hybrid materials, ceramics, glass, shape memory polymers, elastomers and hydrogels. Examples of applications of these materials will be presented, such as in soft robotics, drug delivery systems, 3D electrical circuits, responsive connectors, and medical devices.