Other

The 2nd talk of a monthly webinar series jointly hosted by Perimeter, IVADO, and Institut Courtois. Speaker: David Kremer Garcia, AI Engineer & Lead Data Scientist, IBM Quantum, Yorktown Heights, NY, USA. In this session, I will talk about how we are using AI to improve quantum circuit transpiling and optimization. I will show some of our recent work, where we apply AI methods such as Reinforcement Learning to different transpiling tasks and achieve a remarkable balance between speed and quality of the results. I will also talk about how we integrate these methods with other heuristics to provide "ai-enhanced transpiling" through our Qiskit Transpiler Service.

The Indigenous peoples from the Pacific Ocean are the People of the Sea. Over generations, our ancestors pioneered ways to intentionally travel across thousands of miles of open ocean using only the information nature provides. No external instruments were needed for guidance; rather the navigator became the most sophisticated instrument needed by tuning into the stars, winds, waves, birds and marine life.    In this presentation, I will share these stories along with principles that govern living and operating on traditional voyaging vessels. These values established for the canoes translated directly into the balance is attained between Pacific Indigenous peoples and their island nations. As we envision the future of our planet and look to weave together Indigenous wisdom with western science, we must learn to listen to what is offered by the people of the largest continent on the planet, the Pacific Ocean.

As part of a monthly webinar series jointly hosted by Perimeter, IVADO, and Institut Courtois, Ilya Shpitser will present an introduction to causal inference and its applications to problems in physics and computer science. This seminar will be fully on zoom and members of all three institutes are welcome. 

Abstract: In this talk I will give some history of ideas of causal inference, describe the causal inference workflow, including formalizing the cause-effect question in terms of a parameter, defining (or learning) the causal model, checking if the data has information about the desired parameter via identification theory, and efficiently estimating the parameter if it is identified.  I will briefly touch on connections of causal inference to other areas, discuss what machine learning and causal inference can teach each other, and describe some open problems.   Zoom TBC

Predicting the outcome of scattering processes of elementary particles in colliders is the central achievement of relativistic quantum field theory applied to the fundamental (non-gravitational) interactions of nature. While the gravitational interactions are too minuscule to be observed in the microcosm, they dominate the interactions at large scales. As such the inspiral and merger of black holes and neutron stars in our universe are now routinely observed by gravitational wave detectors. The need for high precision theory predictions of the emitted gravitational waveforms has opened a new window for the application of perturbative quantum field theory techniques to the domain of gravity. In this talk I will show how observables in the classical scattering of black holes and neutron stars can be efficiently computed in a perturbative expansion using a world-line quantum field theory; thereby combining state-of-the-art Feynman integration technology with perturbative quantum gravity. Here, the black holes or neutron stars are modelled as point particles in an effective field theory sense. Fascinatingly, the intrinsic spin of the black holes may be captured by a supersymmetric extension of the world-line theory, enabling the computation of the far field wave-form including  spin and tidal effects to highest precision. I will review our most recent results at the fifth order in the post-Minkowskian expansion amounting to the computations of hundreds of thousands of four loop Feynman integrals.

---

Zoom link 

We develop a Machine-Learning Renormalization Group (MLRG) algorithm to explore and analyze many-body lattice models in statistical physics. Using the representation learning capability of generative modeling, MLRG automatically learns the optimal renormalization group (RG) transformations from self-generated spin configurations and formulates RG equations without human supervision. The algorithm does not focus on simulating any particular lattice model but broadly explores all possible models compatible with the internal and lattice symmetries given the on-site symmetry representation. It can uncover the RG monotone that governs the RG flow, assuming a strong form of the $c$-theorem. This enables several downstream tasks, including unsupervised classification of phases, automatic location of phase transitions or critical points, controlled estimation of critical exponents, and operator scaling dimensions. We demonstrate the MLRG method in two-dimensional lattice models with Ising symmetry and show that the algorithm correctly identifies and characterizes the Ising criticality.

---

Zoom link

Presented in collaboration with Navigating Quantum and AI Career Trajectories: A Beginner’s Mini-Course on Computational Methods and their Applications

---

Deeptech or science-based innovations often spend more than a decade percolating within academic and government labs before their value is recognized (Park et al., 2022). This development lag time prior to venture formation is only partly due to technological development hurdles. Because science-based inventions are often generic in nature (Maine & Garnsey, 2006), meaning that they have broad applicability across many different markets, the problem of identifying a first application requires the confluence of deep technical understanding with expert knowledge of the practice of commercialization. This process of technology-market matching is a critical aspect of the translation of science-based research out of the lab (Pokrajak 2021, Gruber and Tal, 2017; Thomas et al, 2020, Maine et al, 2015) and is often delayed by a lack of capacity to identify, prioritize and protect market opportunities. Typically, deeptech innovations can take 10-15 years of development, and tens (or even hundreds) of millions of dollars of investment to de-risk before a first commercial application (Maine & Seegopaul, 2016). Academics seeking to commercialize such inventions face the daunting challenge of competing for investment dollars in markets that are ill suited to the uncertainty and timescales of deep tech development. The time-money uncertainty challenge faced by science-based innovators is compounded by the fact that most of the scientists and engineers with the world-leading technical skills required to develop science-based inventions, lack innovation skills training, and so cannot navigate the complexities of early and pre-commercialization development critical to venture success. Some researchers, having developed a mix of technical and business expertise, have demonstrated a long-term ability to serially spin out successful ventures (Thomas et al., 2020). Entrepreneurial capabilities, which can be learned, enable scientistentrepreneurs to play formative roles in commercialising lab-based scientific inventions through the formation of well-endowed university spin-offs. (Park et al, 2022; 2024). Commercialization postdocs, when supported by well designed training, stipends, and de-risking supports, can lead the mobilization of fundamental research along multiple commercialization pathways. Recommendations are provided for scholars, practitioners, and policymakers to more effectively commercialise deeptech inventions.

---

Zoom link