Events

Artificial intelligence is rapidly becoming a transformative tool in scientific discovery, playing a crucial role in accelerating research across various fields. In materials science, it has shown the potential to significantly augment traditional methods by providing insights that might have been unattainable using conventional approaches alone. However, the success of machine learning models in materials science is highly dependent on the availability and quality of data. Collecting high-quality datasets remains a major challenge, especially in materials science, where experimental data is often sparse and buried in the unstructured language of scientific literature. AI offers promising solutions for automating data extraction from these vast text resources. Beyond data extraction, it is also being applied to predict material properties, optimize experimental design, and accelerate the discovery of new materials. In this seminar, I will elaborate on how I am integrating AI to address the challenges of data mining from scientific literature. I will also discuss broader AI applications through specific examples, showcasing how AI is reshaping the landscape of materials science.

Prior studies have described the complex interplay that exists between glioma cells and neurons however, the electrophysiological properties endogenous to glioma cells remain obscure. To address this, we employed Patch-sequencing (Patch-seq) on human glioma specimens and found that one-third of patched cells in IDH mutant (IDHmut) tumors demonstrate properties of both neurons and glia. To define these hybrid cells (HCs), which fire single, short action potentials, and discern if they are of tumoral origin, we developed the single cell rule association mining (SCRAM) computational tool to annotate each cell individually. SCRAM revealed that HCs possess select features of GABAergic neurons and oligodendrocyte precursor cells, and include both tumor and non-tumor cells. These studies characterize the combined electrophysiological and molecular properties of human glioma cells and describe a cell type in human glioma with unique electrophysiological and transcriptomic properties that may also exist in the non-tumor brain.

Classification of materials based on their physical properties has shaped the development of modern civilisations. Using the right materials together with fine-tuning of their physical properties, has enabled novel engineered devices. Over the last decade, we have witnessed a new materials classification based on topological concepts. We can group materials using integer numbers associated with the symmetry of their electronic band structure. Recently those innovative concepts have been applied for optical systems to investigate topological phases of light on tailored optical materials. However, examples in the literature are passive optical systems with built-in topology. Active control of topology and device applications remains a challenge. This talk explores the topological control of light across different domains, demonstrating how topological singularities can be harnessed to manipulate electromagnetic waves. First, I will show how exceptional points (EPs) singularities in graphene-based terahertz devices enable precise control over light propagation. By tuning a gate voltage, we can modulate both the intensity and phase of terahertz pulses using a topological phase transition while crossing an EP. We were able to reconstruct the complex energy landscape and explore non-Hermitian physics behind the observed topological properties. In a second approach, we apply topological principles to thermal light. By controlling the reflection topology of thermal emitters, we observed a topological phase transition in thermal radiation by varying a single parameter. The critical point of zero reflection is topologically protected, and boundaries between spatial domains host interface states with near-unity thermal emissivity. Together, these examples illustrate rich physics of topological singularities providing new tools for controlling light, from the terahertz regime to thermal radiation, with potential applications in THz communications and thermal management.

All interested are cordially invited to ICANN & Sabancı University FENS joint webinar on Technical communities make the Internet work and building trust. The webinar will be moderated by Albert Levi and Kürşat Çağıltay and will feature Herv

Linearizability is the standard correctness criteria for implementations of concurrent objects. Roughly speaking, a linearizable implementation of an object gives the illusion that all operations happen sequentially, in some order that respects real-time. Unfortunately, it has been shown that linearizability is expensive to achieve for some concurrent objects, which include useful objects like queues and stacks. Concretely, any linearizable implementation for such objects requires expensive synchronization mechanisms. This negative result limits the scalability of linearizable implementations for some concurrent objects. One way to evade the impossibility result is to relax the semantics of the object to be implemented in order to use only lighter synchronization mechanisms, with the aim of achieving scalability. This talk is about a recently introduced relaxation for queues and stacks called multiplicity. Roughly speaking, multiplicity allows distinct dequeue/pop operations to take the same item only if they are concurrent. It will be shown that queues and stacks with multiplicity can be implemented using only the simplest read/write operations. A variant of multiplicity allows us to develop the fastest relaxed single-enqueuer queue implementations known so far, which has implications for work-stealing, a popular dynamic load balancing technique. This is a joint work with Miguel A Piña, Sergio Rajsbaum and Michel Raynal.

We consider two sequences of holomorphic self maps of the unit disc converging to the boundary, and the corresponding sequences of composition operators acting on the unit ball of H

We introduce several combinatorial properties of three classes of integer partitions: s-modular partitions: A class consisting of partitions into parts with a number of occurrences (i.e., multiplicity) congruent to 0 or 1 modulo s. s-congruent partitions: These generalize Sellers partitions into parts not congruent to 2 modulo 4. s-duplicate partitions: This class includes partitions having distinct odd parts, which are enumerated by the function mypod(n) as a special case. We also present a generalization for Alladis series expansion for the product generating function of mypod(n) and show that Andrews generalization of Göllnitz-Gordon identities coincides with the number of partitions into parts that are simultaneously s-congruent and t-distinct (where parts appear fewer than t times).

Biomanufacturing is a process that involves the use of genetically modified microorganisms to produce a wide range of goods, from pharmaceuticals to biofuels. These microorganisms are engineered to enhance their natural abilities, allowing them to make products with improved characteristics that are more efficient and have less environmental impact than traditional manufacturing methods. We can create a more sustainable and eco-friendly future by harnessing the power of biomanufacturing. Protein-based fibers are produced by many organisms (like silk and wool) by controlling a set of genes. In the lab, genetic engineering can enhance these fiber properties and enable the large-scale production of bioengineered proteins. Our group has adjusted these properties through rational engineering and incorporated new proteins on a 100 kg scale. In this talk, we will discuss technical, financial, and operational challenges on a global scale needed to improve the growth of protein-based biomanufacturing for the textile industry.

This work addresses the novel source separation task of decomposing piano concerto recordings into individual piano and orchestral tracks, enabling pianists at all levels to practice and perform with orchestral accompaniments. As one main contribution, we adapt deep-learning-based source separation techniques, initially designed for the separation of popular music recordings or speech signals. In particular, we address the challenge of higher spectrotemporal correlations between piano and orchestra compared to popular music tracks, and the lack of multitrack datasets for training, by introducing musically motivated data augmentation approaches.