A theoretical understanding of generalization remains an open problem for many machine learning models, including deep networks where overparameterization leads to better performance. Here, we study this problem for kernel regression, which, besides being a popular machine learning method, also describes infinitely overparameterized neural networks. We develop an analytical theory of generalization in kernel regression using the replica method of statistical mechanics. This theory is applicable to any kernel and data distribution. Experiments with practical kernels, including those arising from wide neural networks, show perfect agreement with our theory. Further, our theory accurately predicts the generalization performance of neural networks with modest widths. We provide an in-depth analysis of our analytical expression for kernel generalization. We show that kernel machines employ an inductive bias towards simple functions, preventing them from overfitting the data. We characterize whether a kernel is compatible with a learning task in terms of sample efficiency. We identify a first-order phase transition in our theory where more data may impair generalization when the task is noisy or not expressible by the kernel. Finally, we extend these results to out-of-distribution generalization.

The Inverse Galois problem asks whether every finite group G is the Galois group of some Galois extension of the field of rational numbers Q. The Malle conjecture is a quantitative version of it: it predicts the number of Galois extensions of Q with the Galois group G of bounded size (such as the discriminant). In this talk we will present how one can generalize the conjecture for finite group schemes. We show how the generalization helps explain inconsistencies of the Malle conjecture found by Klüners. The talk is based on a joint work with Takehiko Yasuda.

Distributed systems are prone to concurrency bugs due to the nondeterminism in the interleavings of the concurrent events in an execution. Detecting and diagnosing concurrency bugs in distributed systems is critical since unforeseen interleavings of concurrent messages, network, or process faults can result in unexpected, erroneous system behavior. However, concurrency bugs are hard to detect as they are triggered only in some subtle interleavings of the events. Random testing is a practical way of searching for bugs in large distributed systems. While na

Abiotic stresses, such as drought, salinity, extreme temperatures, and heavy metal toxicity, significantly constrain crop productivity and food security worldwide. Seaweed-based biostimulants, largely manufactured using brown and red seaweeds, have emerged as a tool to mitigate abiotic stresses. These biostimulants enhance plant resilience by modulating physiological and molecular mechanisms. The bioactive compounds such as polysaccharides, polyphenols, and secondary metabolites present in seaweed biostimulants affect stress-responsive pathways, enhancing photosynthetic efficiency, water-use efficiency, and cellular homeostasis. At a molecular level, the bioactive molecules affect transcriptional and post-translational processes involved in the regulation of gene expression leading to better plant performance under abiotic stresses. The presentation will include result of studies conducted at the Marine bioproducts laboratory, Dalhousie University.

Cellular therapies using engineered cells are triggering a paradigm shift in the development of biopharmaceuticals. Instead of using simply a single protein (e.g. antibody, human proteins etc), microbial therapeutics enables us to design biological machines with multifunctional properties. Microbial therapeutics are mainly designed as three modules, including: sensing- targeting and delivery. In our laboratory we have been developing microbial therapeutics for various diseases including Type1 diabetes, breast/column cancer, anemia and rare metabolic diseases. We have established various cellular protein secretion systems for the delivery of the biotechnological drugs. In addition to logic gated cell surface systems, recently we have developed a complex genetic circuitry to control the permeability/leakiness of the cell membrane in a probiotic bacterium E. coli Nissle 1917 for payload delivery of protein drugs. I will give an overview of our recent research efforts focused but not limited only with living drugs.

Abstract: Human-computer interaction (HCI) can be an active or passive source of data on human behavior, which, when analyzed over time and/or across cohorts, contains clues about our cognitive and mental health. For example, if we capture speed of movement / gait from smartphones, eye movements from mobile or stationary cameras, skin conductivity from smart watches, and/or visuo-spatial memory errors from virtual reality experiments, we get a set of digital biomarkers of cognitive health. In this talk, we examine how digital biomarkers are used in various contexts, enriched by examples from empirical research outcomes from by Dr. Çöltekin and her team.

Plant biotechnology has emerged as a pivotal tool in agriculture and industry, enhancing crop resilience, productivity and nutritional value while also offering sustainable solutions for industrial applications. Techniques such as genetic modification, genome editing, marker-assisted selection and omics technologies facilitate the development of plants with improved biotic and abiotic stress tolerance, including resistance to diseases, drought and salinity. These advancements contribute to food security and sustainable agriculture practices, addressing global challenges like climate change, population growth, and resource scarcity. Additionally, plant tissue culture and cell suspension culture enable large-scale production of valuable plant species, plant-derived extracellular vesicles and bioactive molecules, including secondary metabolites with applications in pharmaceuticals, cosmetics and the food industry. Through micropropagation, rapid clonal propagation of elite genotypes and pathogen-free plants are achieved, ensuring uniformity and enhanced yield in crops. Conversely, conserving the gene pool is crucial for advancements in these fields, beyond just achieving uniformity. Therefore, both tissue culture and DNA barcoding play vital roles in supporting biodiversity and conservation efforts. Collectively, these technologies offer transformative solutions to meet global agricultural demands and open new industrial applications, positioning plant biotechnology as a cornerstone of sustainable development. This talk will explore our research into both agricultural and non-agricultural applications of plant biotechnology based on these powerful tools.

Abstract: In product design, when we need long-term use, high thermal stability, high strength, and resistance to chemicals, we usually end up in a materials selection region that is covered by ceramics. On the other hand, these properties that single out ceramics for extreme applications are obstacles to conventional manufacturing methods, and most of the time, the uncomfortable outcome is settling for either a sub-optimal material or shape. At present, we do not have cost- and energy-effective, standard processing routes for ceramics that utilize current setups for rapid prototyping, and manufacturing in low numbers. There is a clear need for alternative methods that are specifically designed for ceramics. We generate self-standing doughs of advanced ceramics that can be shaped by hand, and via traditional and laser machining at the green state. We design particle-specific polymeric additives that homogeneously coagulate ceramic suspensions through polymer bridging, imparting a clay-like rheology to the suspensions of zirconia, alumina, magnesia, and many other technologically relevant ceramic systems. Starting from a self-standing structure that can withstand machining forces enables the exploitation of existing setups for machining of metals. The additive content in these formulations is less than 2 wt. the formulation can be reused, and the products reach 98 of the theoretical density after sintering. Recently, we have expanded our materials portfolio to nitrides (SiAlONs, more specifically), and our preliminary results on carbides and metals are promising. Large-scale imprinting and injection molding are also applicable. This

Abstract: In product design, when we need long-term use, high thermal stability, high strength, and resistance to chemicals, we usually end up in a materials selection region that is covered by ceramics. On the other hand, these properties that single out ceramics for extreme applications are obstacles to conventional manufacturing methods, and most of the time, the uncomfortable outcome is settling for either a sub-optimal material or shape. At present, we do not have cost- and energy-effective, standard processing routes for ceramics that utilize current setups for rapid prototyping, and manufacturing in low numbers. There is a clear need for alternative methods that are specifically designed for ceramics. We generate self-standing doughs of advanced ceramics that can be shaped by hand, and via traditional and laser machining at the green state. We design particle-specific polymeric additives that homogeneously coagulate ceramic suspensions through polymer bridging, imparting a clay-like rheology to the suspensions of zirconia, alumina, magnesia, and many other technologically relevant ceramic systems. Starting from a self-standing structure that can withstand machining forces enables the exploitation of existing setups for machining of metals. The additive content in these formulations is less than 2 wt. the formulation can be reused, and the products reach 98 of the theoretical density after sintering. Recently, we have expanded our materials portfolio to nitrides (SiAlONs, more specifically), and our preliminary results on carbides and metals are promising. Large-scale imprinting and injection molding are also applicable. This

Abstract: The Uber Eats home feed recommendation system aims to provide a magical browsing experience to our 150M+ monthly active customers. To this end, the recommendation system leverages machine learning to create a personalized list of stores every time users open the app. In this talk, we will discuss the evolution of this recommendation system. Additionally, we will deep dive into some interesting real life problems that we have encountered in the past, and finally discuss competencies typically required for ML developer/applied scientist roles in the industry.