Lifeboat Foundation: Safeguarding Humanity
Toggle light / dark theme

Blog

Category: quantum physics

Hierarchical Prototyoe Emergence in Hopfield Networks

How can an AI memory learn the patterns behind the patterns?

A new study, Hierarchical Prototype Emergence in Modern Hopfield Models, explores how associative-memory models can learn not just individual examples, but the deeper structure connecting them—and potentially use that structure to create new, sensible examples. https://arxiv.org/abs/2609.

The researchers study a modern version of a Hopfield network, an AI system designed to store and retrieve patterns. Instead of treating every memory as unrelated, they organize the memories into a hierarchy, much like a family tree: individual images belong to groups, groups share broader characteristics, and those groups may themselves belong to larger categories.

The key question is whether the network can go beyond simply remembering the images it was given. The researchers find that, under certain conditions, the network develops prototypes—stable representations that capture common features shared by multiple memories. In other words, rather than memorizing a particular example, the system can discover something like the “essence” of a group and reconstruct a new example from it.

Interestingly, the amount of information needed to achieve this kind of generalization grows only quasi-polynomially with the complexity of the hierarchy, suggesting that learning these higher-level patterns may be considerably more efficient than simply memorizing every possible combination.

The researchers also find similar behavior when testing Fashion-MNIST data: the transition between memorization and prototype formation depends on factors such as how many memories the network stores and the sharpness of its activation function.

The broader significance is that this provides a mathematical “toy model” for a much bigger question in AI: how can a system move from memorizing examples to learning hierarchical structure and generating something genuinely new from that structure? Understanding this process in Hopfield models could offer clues about more sophisticated generative architectures, including diffusion models.

This is especially interesting because it frames generalization as the emergence of stable higher-level representations, rather than simply as better recall of training examples.

Continue reading “Hierarchical Prototyoe Emergence in Hopfield Networks” | >

Heavy fermions emerge at an atomic-layer interface, unlocking new ways to design quantum materials

A research team led by the University of Osaka has directly observed, for the first time, an unusual heavy-fermion state forming at the boundary between a one-atom-thick material and a metal. Such states are closely linked to exotic quantum phenomena, including unconventional superconductivity, and the finding opens new possibilities for designing quantum materials through their interfaces.

The researchers created a high-quality, one-atom-thick layer of ytterbium–copper (YbCu₂) on a copper crystal and examined how electrons behaved across the interface using intense synchrotron light. Their measurements showed that electrons localized in the atomic layer interact with mobile electrons in the underlying copper to form the heavy-fermion state.

The measurements revealed two distinct heavy-fermion states. One was confined mainly to the two-dimensional YbCu₂ layer, while the other extended into the three-dimensional copper substrate. Crucially, the latter arose from hybridization between localized Yb 4f electrons in the atomic layer and mobile conduction electrons in the underlying copper, providing direct evidence of an interfacial heavy-fermion state.

X-ray technique reveals how quantum materials respond to laser pulses in real time

Creating a quantum device often begins by intentionally damaging a crystal. Scientists fire an ultrafast laser pulse into a material, knocking atoms out of place and leaving behind tiny imperfections called vacancies. Far from being flaws, these vacancies can behave as qubits—the fundamental building blocks of quantum information.

The challenge is precision.

Researchers want to place these quantum defects exactly where they need them, but until now they have had only a limited understanding of what happens inside a crystal during the instant when a laser creates the defect.

Can a passing black hole disturbance leave a trace in Hawking radiation?

Black holes, despite their extreme nature, are described by surprisingly few basic parameters: their mass, charge and spin. Two black holes may have formed for completely different reasons and passed through very different histories, yet still end up in the same final state. John Wheeler summarized this idea with the famous phrase “black holes have no hair.”

Once quantum physics enters the picture, things become more complicated. As Stephen Hawking showed in the 1970s, black holes have a temperature and produce what we now call Hawking radiation. For a stationary black hole, this radiation has a characteristic temperature related to the surface gravity.

The real universe is much less ideal. Finding a completely isolated black hole is not easy. Matter may fall toward it, a nearby companion may disturb the surrounding spacetime, or external fields may change. For some period of time, the black hole can therefore be away from its quiet stationary state.

Reading hidden topology in light, even when energy leaks away

When I explain topology to students, I start with a knot in a rope. You can stretch it, twist it or shake it, but the knot stays until you cut the rope. Physicists have found that some materials and devices carry similar “knots” in how waves move through them. These are topological properties, labeled by whole numbers that don’t change under small imperfections. That robustness is why topology has become one of the central ideas in modern physics. It promises electronics, photonics and quantum devices that tolerate defects and noise.

There is a catch that has always bothered us. These topological numbers live in what physicists call momentum space. It’s an abstract space that describes how a wave travels, not where it is.

In most experiments, nobody looks at momentum space directly. Instead, we infer the topology from its consequences, such as special states appearing at the edges of a carefully fabricated sample. That works, but it is a bit like working out whether a rope is knotted by looking only at its ends.

Scientists uncover recurrent patterns within chaotic quantum behavior

Many complex quantum systems rapidly lose the recognizable patterns of their initial states as their components interact. To describe patterns of regular and chaotic motion in specific systems, physicists can construct a mathematical map called an effective phase space.

In some classical systems that follow familiar laws of motion, orderly and chaotic paths occupy different regions of phase space. Yet establishing whether a comparable pattern exists in quantum many-body systems (i.e., systems with many interacting quantum components) has so far proved challenging, partly because interactions in these systems can produce a quantum phenomenon called entanglement.

When parts of a system are entangled, their combined quantum state cannot be fully described by treating each independently.

Quantum interactions may have locked early universe’s fields into existing energy states

The universe may be trapped in its own comfort zone, and a researcher in the College of Engineering and Computer Science has helped explain why it cannot seem to leave. A new study suggests the universe could be locked into its current state by the same kinds of quantum effects that scientists study when trying to preserve fragile information inside a quantum computer.

The paper, titled “Cosmic Lockdown: When Decoherence Saves the Universe from Tunneling,” has been accepted for publication in the Journal of Cosmology and Astroparticle Physics.

It was written by Gregory Kaplanek, a postdoctoral researcher working in the lab of Jason Pollack, an assistant professor in the Department of Electrical Engineering and Computer Science, along with four collaborators: Robson Christie and Jaewoo Joo of the University of Portsmouth in the United Kingdom, Vincent Vennin of the Laboratoire de Physique de l’Ecole Normale Superieure in France and David Wands, also of Portsmouth. Kaplanek and Pollack are part of Syracuse University’s Institute for Quantum and Information Sciences.

Quantum systems never quite forget where they came from

Even the most chaotic quantum systems keep a permanent mark of their own past—a “quantum birthmark”—that never fades. Researchers from Tampere University, Harvard University and TU Dresden discovered the feature in their recent study. Their findings shed new light on the elusive relationship between classical and quantum mechanics. Looking ahead, these quantum birthmarks, along with related phenomena known as “scars,” could eventually be harnessed to power next-generation nanoelectronics.

Stir a drop of milk into a cup of tea and it is gone for good. The swirl blurs, spreads and evens out, and no amount of staring at the cup will tell you where the drop first landed. Physicists refer to this kind of unpredictable, memory-erasing behavior simply as chaos, and it underpins much of how we explain the everyday world: heat spreading through a room, smoke filling the air, a pinball rattling away from wherever it was launched.

The new study shows that the quantum realm—the world of atoms and electrons—does not play by that rule. There, the earliest moments leave a signature that never washes out. The study “Quantum Birthmarks: Ergodicity Breaking Beyond Scarring” was published in Physical Review X on Sept. 10, 2026.

Two Laser Pulses Unlock Electron States Physicists Had Never Reached Before

Two ultrashort laser pulses crossed in space have unlocked previously inaccessible electron states.

At the point where two ultrashort laser pulses cross, physicists can now make light oscillate through all three dimensions of space. That unusual geometry matters because some electronic quantum states have remained out of reach in experiments, even when theory said they should exist. By shaping and overlapping laser pulses from different directions, researchers at the University of Oldenburg in Germany have now created those three-dimensional light fields and used them to reach previously inaccessible states of electrons.

To build the fields, the team shaped two femtosecond laser pulses, each lasting only a few millionths of a billionth of a second, and sent them toward the same point from different directions. The beams also carried different colors of light. Where they intersected, their electric fields combined into a three-dimensional structure whose shape the researchers could control.

Tail-Likelihood Reinforcement Learning: Teaching AI to Keep Its Best Possibilities Alive

Most reinforcement learning methods train an AI to maximize its average reward. That sounds sensible—but averages can hide something important. https://arxiv.org/abs/2609.

Think of two AI systems that both score an average of 70. One almost always produces results around 70. The other usually scores 60, but occasionally discovers an exceptional solution scoring 100. If we are allowed to sample the AI many times and choose the best result, the second system could be far more valuable.

This is the problem addressed by Tail-Likelihood Reinforcement Learning (TailRL).

Instead of asking only, “What reward does the AI achieve on average?”, TailRL asks a different question across the entire range of outcomes:

“How likely is the AI to produce something better than a particular reward threshold?”

The method effectively turns continuous rewards into a collection of success-or-failure questions. Importantly, it gives greater learning weight to rare, high-reward outcomes rather than allowing them to disappear into the average.

There is an interesting connection to Best-of-k sampling: when we generate multiple attempts and keep the best one, performance depends on whether the model has retained enough probability mass around those unusually good solutions. TailRL explicitly trains the model to preserve that potential.

The practical advantage is significant. The researchers report that TailRL helps models avoid getting trapped in mediocre solutions across tasks including object localization, maze navigation, GUI grounding, and code optimization. More importantly, the resulting models become better at exploiting additional sampling at inference time—the more attempts you give them, the more useful those attempts become.

Continue reading “Tail-Likelihood Reinforcement Learning: Teaching AI to Keep Its Best Possibilities Alive” | >
/* */