Tuesday, February 17, 2026
Alioscia Hamma (Universita di Napoli Federico II)

Title: To learn a mocking-black hole 

The problem of decoding information scrambled by a black hole lies at the heart of the black hole information paradox: semiclassical Hawking radiation appears thermal and featureless, suggesting information loss, yet quantum theory mandates unitarity and information preservation. In this talk, I will explore a quantum information-theoretic approach to this challenge inspired by the Hayden-Preskill-Yoshida-Kitaev protocol. We show that even for “quasi-chaotic” scramblers — unitaries that mimic the internal dynamics of black holes — it is possible to construct efficient decoders without prior detailed knowledge of the scrambler. This result challenges the conventional intuition that chaotic scrambling irrevocably hides information behind exponential complexity barriers, as conjectured in holographic models of black holes. Moreover, we prove something impossible: unscrambling of complex behavior can be achieved by a simple process. 

=====================================================

Tuesday, March 3, 2026
Akira Tanji (Keio University, Japan)

Title: Learning from Quantum Data with Limited Copies and Imperfections

Learning from quantum data is constrained by limited copies of quantum states, trainability issues, and imperfections in realistic quantum systems. In this talk, I present learning algorithms designed to address these practical limitations. First, I present a framework for quantum phase classification via partial tomography-based quantum hypothesis testing. Using classical shadows, a standard partial-tomography technique, we estimate reduced density matrices and construct localized Helstrom measurements, enabling accurate classification while reducing the required number of state copies. Second, I describe learning from imperfect quantum data via unsupervised domain adaptation on classical shadows. By combining shadow-based features with domain-adversarial learning, the method transfers knowledge from ideal states to noisy or algorithmically biased ones, enabling reliable predictions under hardware noise and algorithmic errors. These results advance quantum data analysis toward practical applications in realistic quantum settings. 

=====================================================

Thursday, March 12, 2026
Alioscia Hamma (Universita di Napoli Federico II)

Title: QTris, a game for Quantum Mechanics


One of the major obstacles in the teaching and understanding of quantum mechanics is the interpretation of its rules and its main constructs, such as the wave-function. One can take seriously the idea that understanding means to understand how to use the rules of the game. QTris is a boardgame for the gentle introduction and teaching of quantum mechanics from elementary to graduate school. I even use it to remind people in my group what quantum mecanichs is about. 

=====================================================

Tuesday, March 31, 2026
Ryotaro Okabe (MIT)

Title: Accelerating Quantum Material Discovery with Machine Learning: 

From Property Prediction to Materials Design

 Materials are the foundation for solving nearly every major technological challenge of our time — including quantum computing, secure communication, clean energy, medical imaging, and next-generation electronics. Quantum materials, including topological insulators, frustrated magnets, and unconventional superconductors, sit at the frontier of these applications, but discovering them is extraordinarily hard. The space of possible crystal structures is astronomically large, and each candidate requires costly first-principles calculations or slow experimental synthesis. In this talk, I will describe a research program built on a central idea: machine learning guided by the physical symmetry of crystal structures is more accurate, data-efficient, and physically interpretable than black-box alternatives. I will present work on predicting phonon dispersions and optical spectra directly from crystal structure, and on a generative AI framework that produces millions of candidate quantum materials filtered for stability, novelty, and experimental viability. I will also discuss how AI models are beginning to assist with synthesis planning, and what an AI-driven discovery loop for quantum materials could look like in the near term. 

=====================================================

Tuesday, April 7, 2026
No colloquium: Departmental Meeting 

=====================================================

Tuesday, April 14, 2026
Chuteng Zhou (Arms)

Model Hardware Co-design in the Age of AI

Artificial Intelligence (AI) is transforming society, but its rapid growth comes with increasing demands on energy and computational resources. At the same time, it presents a unique opportunity to rethink the computing stack. A key emerging trend is the tighter coupling between machine learning models and hardware: models are increasingly designed to match the characteristics of target hardware, while new hardware architectures are developed with direct feedback from model behavior. This co-design paradigm is becoming increasingly important for sustaining AI scaling. In this talk, I will present recent industry research on model-hardware co-design, including approaches for analog Compute-in-Memory (CiM), model optimization for edge Neural Processing Units (NPUs) and algorithm-hardware co-design for low-precision federated learning. I will argue that closer integration between models and hardware is critical to unlocking the next generation of efficient, scalable and decentralized AI systems.

=====================================================

Tuesday, April 21, 2026
Berthy Feng (MIT)

Imaging at the Edge of Science: Integrating Scientific Knowledge and AI to Recover Hidden Structure

Images play a central role in scientific discovery. Whether it’s astronomical, biological, or materials systems, bringing complex phenomena into view enables scientists to probe, model, and fundamentally understand them. However, many of the most important scientific questions lie at the edge of what can be directly observed. We can accomplish extreme imaging through computational methods, bringing the invisible into view by supplementing limited observable data with human-imposed assumptions, or priors. When imaging for science, the challenge is imposing just enough known assumptions to infer the unknown. I create principled methods for bringing advanced priors, such as scientific knowledge and AI, into computational imaging. Using astrophysics as a running example, this talk presents my vision for a framework in which scientists systematically explore different priors, understand their effects on imaging, and extract scientific insights.

The talk is organized in three parts.

1. First, we understand the importance of priors in extreme scientific imaging. I present my work on leveraging generative AI to flexibly tune a knob between different priors and understand their effects on imaging. Applied to black-hole imaging, my approach lets us infer physical features of a real black hole by identifying image features that are robust to prior assumptions.

2. Second, we carefully balance scientific assumptions to solve an extreme imaging problem in astrophysics. I present Physics-informed Dynamic Emission Fields (PI-DEF), a method for imaging the dynamic 3D gas near a black hole. PI-DEF strikes a balance between known/unknown physics, imposing known physics as hard constraints on the solution while leaving room for learning unknown physics, such as the velocity field near the black hole.

3. Third, we open an efficient route for bringing in known physics across imaging problems. I present Neural Approximate Mirror Maps (NAMMs), which learn to automatically impose any desired physics constraint onto any image. With NAMMs, we can easily incorporate known constraints (e.g., conservation laws) into generated and reconstructed images.

The ideas of my talk naturally extend to many scientific domains, including biology, chemistry, and materials science.

=====================================================

Tuesday, April 28, 2026
Matthew Weiss (UMass Boston)

QBism Goes to Google 

At the heart of quantum mechanics is the Born rule, used to calculate probabilities for experimental outcomes. It is usually formulated in terms of mathematical objects such as wave functions, density matrices, projectors, POVM elements, and so on. QBism offers an alternative perspective: the Born rule is best understood as a way of checking the consistency of one's probability judgments with respect to a reference measurement. In particular, the Born rule can be rewritten as a deformation of the usual law of total probability, and it has been known for some years that so-called SIC measurements minimize this deformation, making the quantum rule look as classical as possible. At the same time, the study of nonstabilizerness, colloquially known as "magic," has emerged as a powerful way of quantifying not only the usefulness of certain quantum states for quantum computation but also the difficulty of preparing them. Remarkably, the very same SIC states have maximal magic, leading to a beautiful seeming paradox: the most magical states make quantum theory look as classical as possible. Recently, the QBism group has partnered with Michel Devoret's group at Google Quantum AI to implement SIC measurements on their Willow quantum processor, using the stringent consistency conditions as a stress test for their hardware. I will give a sense of how SIC measurements may be performed on qubit based quantum computers, what battery of experiments we are running, and offer several preliminary results of our collaboration.

=====================================================

Tuesday, May 5, 2026
No colloquium: Departmental Meeting 

=====================================================

Tuesday, May 12, 2026
Liang Fu (MIT)

AI from First-Principles: a Foundation Model for Quantum Matter

The properties of all materials—from semiconductors to superconductors—ultimately emerge from quantum mechanics at the microscopic scale. Can AI learn directly from the laws of physics without training data and predict material properties? In this talk, I will present a neural-network approach that variationally solves the many-electron Schrödinger equation and uncovers diverse quantum phases of matter. Using a single, provably universal Fermi architecture, this approach has led to the discovery of a quantum electron quasicrystal and the crystallization of fractional quantum Hall liquids. I will also discuss the prospect of a Large Electron Model: a foundation model for matter, with applications spanning molecules, materials, and quantum hardware.

=====================================================