Search Talks

In this talk, I will argue and practically illustrate that insights in quantum information, concretely coming from the tensor network representations of quantum many-body states, can help in devising better privacy-preserving machine learning algorithms. In the first part, I will show that standard neural networks are vulnerable to a type of privacy leak that involves global properties of the data used for training, thus being a priori resistant to standard protection mechanisms. In the second, I will show that tensor networks, when used as machine learning architectures, are invulnerable to this vulnerability. The proof of the resilience is based on the existence of canonical forms for such architectures. Given the growing expertise in training tensor networks and the recent interest in tensor-based reformulations of popular machine learning architectures, these results imply that one may not have to be forced to make a choice between accuracy in prediction and ensuring the privacy of the information processed when using machine learning on sensitive data.

---

Zoom link

Causal discovery aims at learning causal relations among variables from data. It is an emerging field at the interface of machine learning and statistics that found ample application in several disciplines. From a physicist's point of view it can be seen as an operational definition of the concept of causal relation between variables, especially in an observational context where experimental manipulation is precluded. Interestingly, causal discovery has not yet been applied to Astronomy, despite it being the observational science par excellence. Here I will present the first application of causal discovery to Astronomy with the goal of addressing a debated issue in galaxy formation: the origin of the observed scaling relations between supermassive black hole (SMBH) and host galaxy properties. I apply three causal discovery algorithms to a state-of-the-art dataset of SMBH host galaxies with dynamical mass measurements, with the goal of learning a causal structure in terms of a directed acyclic graph. The results are consistent between methods and are amenable to physical interpretation, showing that across the multiplicity of possible causal structures, in elliptical galaxies SMBH mass is predominantly an effect of galaxy properties while in spiral galaxies the reverse holds. I offer an explanation of this finding in terms of the physics of galaxy merging, and address the limitations and the theoretical implications of this new method for galaxy formation in some detail.

---

Zoom link TBA

Inflation remains one of the enigmas in fundamental physics. While it is difficult to distinguish different inflation models, information contained in primordial non-Gaussianity (PNG) offers a route to break the degeneracy. In galaxy surveys, the local type PNG is usually probed by measuring the scale-dependent bias in the galaxy power spectrum on large scales, where cosmic variance and systematics are also large. Other types of PNG need bispectrum, which is computationally challenging and is contaminated by gravity. I will introduce a new approach to measuring PNG by using the reconstructed density field, a density field reversed to the initial conditions from late time. With the reconstructed density field, we can fit a new template at the field level, or compute a near optimal bispectrum estimator, to constrain PNG. By reconstructing the initial conditions, we remove the nonlinearity induced by gravity, which is a source of confusion when measuring PNG. Near optimal bispectrum estimator mitigates computational challenges. This new approach shows strong constraining power, offers an alternative way to the existing method with different systematics, and also follows organically the procedure of baryon acoustic oscillation (BAO) analysis in large galaxy surveys. I will present a reconstruction method using convolutional neural networks that significantly improves the performance of traditional reconstruction algorithms in the matter density field, which is crucial for more effectively probing PNG. This pipeline can enable new observational constraints on PNG from the ongoing Dark Energy Spectroscopic Instrument (DESI) and Euclid surveys, as well as from upcoming surveys, such as that of the Nancy Grace Roman Space Telescope.

---

Zoom link https://pitp.zoom.us/j/92361466496?pwd=ZlljUGlKaTVlSFZIV21NUHNGY2RRUT09

I will begin by reviewing the general mathematical concept of representation. I will then show that representation theory is more generally applicable than one might expect, for example in quantum foundations, in quantum gravity and in machine learning. In those contexts, I will first talk about the notion of representation underlying phenomena of emergence in quantum gravity. I will then discuss how quantum reference frames might be viewed as representations. Finally, I will talk about how transformer models, such as GPT-4, construct representations of what they learn and, in that light, what it may take for machines to reach AGI or even consciousness.  

---

Zoom link: https://pitp.zoom.us/j/99349397588?pwd=T056VjZXRWZWT28zY3VOZHFDcmQ3QT09

With rapid progress in simulation of strongly interacting quantum Hamiltonians, the challenge in characterizing unknown phases becomes a bottleneck for scientific progress. We demonstrate that a Quantum-Classical hybrid approach (QuCl) of mining the projective snapshots with interpretable classical machine learning, can unveil new signatures of seemingly featureless quantum states. The Kitaev-Heisenberg model on a honeycomb lattice with bond-dependent frustrated interactions presents an ideal system to test QuCl. The model hosts a wealth of quantum spin liquid states: gapped and gapless Z2 spin liquids, and a chiral spin liquid (CSL) phase in a small external magnetic field. Recently, various simulations have found a new intermediate gapless phase (IGP), sandwiched between the CSL and a partially polarized phase, launching a debate over its elusive nature. We reveal signatures of phases in the model by contrasting two phases pairwise using an interpretable neural network, the correlator convolutional neural network (CCNN). We train the CCNN with a labeled collection of sampled projective measurements and reveal signatures of each phase through regularization path analysis. We show that QuCl reproduces known features of established spin liquid phases and ordered phases. Most significantly, we identify a signature motif of the field-induced IGP in the spin channel perpendicular to the field direction, which we interpret as a signature of Friedel oscillations of gapless spinons forming a Fermi surface. Our predictions can guide future experimental searches for U(1) spin liquids.

--

Zoom link: https://pitp.zoom.us/j/94233944575?pwd=OVljLzMrZzlKeUErNHZQRkEzMFRKUT09

While General Relativity has withstood tests on solar system scales, progress in observational cosmology now enables tests on the largest scales. I will present results from our tests of gravity using Dark Energy Survey Year 3 weak lensing and clustering data in addition to a variety of complementary data. One outcome of this analysis was the necessity to further explore lensing consistency so I will present preliminary results of such tests using the latest Cosmic Microwave Background (CMB) lensing data. These analyses are setting the scene for future tests of fundamental physics with CMB and galaxy surveys: I will show expected results and challenges from the Rubin Observatory. I will finally argue for the use of machine learning for theory exploration, to better organize our efforts within the future experimental landscape. As PI is a leading center in outreach, I will end my talk by sharing my experience with science content creation on various social media platforms.

Zoom link: https://pitp.zoom.us/j/96929387143?pwd=WW1CZElpMkN2U1RQUjB3VERxRVQ5dz09

In 2020 our Oxford-based Quantinuum team performed Quantum Natural Language Processing (QNLP) on IBM quantum hardware [1, 2].  Key to having been able to achieve what is conceived as a heavily data-driven task, is the observation that quantum theory and natural language are governed by much of the same compositional structure -- a.k.a. tensor structure.  

Hence our language model is in a sense quantum-native, and we provide an analogy with simulation of quantum systems in terms of algorithmic speed-up [forthcoming].  Meanwhile we have made all our software available open-source, and with support [github.com/CQCL/lambeq].    

The compositional match between natural language and quantum extends to other domains than language, and argue that a new generation of AI can emerge when fully pushing this analogy, while exploiting the completeness of categorical quantum mechanics / ZX-calculus [3, 4, 5] for novel reasoning purposes that go hand-in-hand with modern machine learning.

[1]     B. Coecke, G. De Felice, K. Meichanetzidis and A. Toumi (2020) Foundations for Near-Term Quantum Natural Language Processing.  https://arxiv.org/abs/2012.03755

[2]     R. Lorenz, A. Pearson, K. Meichanetzidis, D. Kartsaklis and B. Coecke (2020) QNLP in Practice: Running Compositional Models of Meaning on a Quantum Computer.  https://arxiv.org/abs/2102.12846

[3]     B. Coecke and A. Kissinger (2017) Picturing Quantum Processes. A first course on quantum theory and diagrammatic reasoning.  Cambridge University Press.

[4]     B. Coecke, D. Horsman, A. Kissinger and Q. Wang (2021) Kindergarten quantum mechanics graduates (...or how I learned to stop gluing LEGO together and love the ZX-calculus).  https://arxiv.org/abs/2102.10984

[5]     B. Coecke and S. Gogioso (2022) Quantum in Pictures.  Quantinuum, 2023.

Zoom Link: https://pitp.zoom.us/j/92333285960?pwd=MlpJSklmMlVlUlRTTWhsNjc2T2Y4QT09