Pattern Recognition Lab

The HydroGym reinforcement learning platform for fluid dynamics

For decades, the art of controlling how fluids move—from the air flowing over an airplane wing to the water rushing through a pipe—has been a stubbornly difficult challenge. Fluid dynamics, the study of everything from gentle breezes to powerful hurricanes, is defined by intricate, high-dimensional, and non-linear behavior. To tame these turbulent waters, researchers traditionally had to treat every single problem—every new wing shape, every unique airflow condition—as a unique puzzle. This meant that when a control strategy worked brilliantly for one scenario, it often crumbled when applied to the next, trapping progress in a sea of isolated case studies.

Plugging into the Flow: How a Digital Gym is Revolutionizing Aerodynamics For decades, the art of controlling how fluids move—from the air flowing over an airplane wing to the water rushing through a pipe—has been a stubbornly difficult challenge.

Why we still need software engineering — Anthropic on patterns and problems in emerging multiagent systems

The march toward Artificial General Intelligence—creating systems capable of performing any intellectual task a human can—is no longer a distant science fiction fantasy. It is a rapidly accelerating engineering challenge. We are moving from crafting brilliant, singular digital minds to orchestrating sprawling, complex digital societies where numerous AI agents interact, debate, and build together. It feels inevitable, perhaps even magical, that superior intelligence will simply *organize* itself.

When Agents Collide: Why Software Engineering Still Holds the Keys to AGI's Future The march toward Artificial General Intelligence—creating systems capable of performing any intellectual task a human can—is no longer a distant science fiction fantasy.

How to build an AI-driven digital organism

Biology, the intricate, magnificent language of life, has long resisted neat encapsulation. It is a realm defined by staggering complexity: a single gene mutation can cascade through complex cellular signaling, alter the function of an entire organ, and ultimately dictate a patient's health outcome. In the physical world, manipulating this machinery is often slow, incredibly expensive, and fraught with high risk. But what if we could bypass the expensive wet-lab bottlenecks? What if we could build a perfect, safe, and high-throughput digital sandbox where we could design, test, and reprogram life itself?

From Code to Cell: How Scientists Plan to Build Life in a Digital Organism Biology, the intricate, magnificent language of life, has long resisted neat encapsulation.

Structural and dynamical strategies to prevent runaway excitation in reservoir computing

In the dazzling world of artificial intelligence, where machines are learning to mimic the intricate timing and complexity of the human brain, a particular architecture stands out: Reservoir Computing (RC). Imagine a complex, randomized network—a 'reservoir'—that acts as a dynamic, internal memory bank. Unlike traditional computer programs that move data strictly from point A to point B, an RC allows information to flow in circles, letting the network maintain a self-sustaining, rich tapestry of internal activity. This ability to hold and process complex temporal patterns is what makes RC so tantalizing for applications ranging from speech recognition to natural language understanding.

Keeping the Engine from Redlining: Taming Neural Overdrive in Smart Computing In the dazzling world of artificial intelligence, where machines are learning to mimic the intricate timing and complexity of the human brain, a particular architecture stands out: Reservoir Computing (RC).

Physics-informed digital twin and onboard control of a brainbot for intelligent active matter

In the vast and exhilarating landscape of modern physics, some frontiers are less about grand cosmic forces and more about the exquisite dance of matter at the microscopic level. We are witnessing a revolution in how we view movement itself—the emergence of what scientists call “active matter.” These are materials that possess the uncanny ability to move, self-organize, and interact, often driven not by external wind or current, but by their own internal energy. Think of swarms of microscopic robots, bacteria swimming in a complex fluid, or self-propelled particles navigating dense environments. For decades, these systems have fascinated researchers, presenting a rich playground for understanding nonequilibrium physics. However, bringing these dynamic entities to life with genuine intelligence—making them truly adaptive agents—has remained a monumental challenge.

Giving the Bristlebot a Brain: Building Self-Aware Micro-Machines in the Age of Active Matter In the vast and exhilarating landscape of modern physics, some frontiers are less about grand cosmic forces and more about the exquisite dance of matter at the microscopic level.

From Noise to Precision: Momentum-Driven Adaptive DC Offset Calibration for Enhanced Blood Pressure Displacement Extraction Using FMCW Radars

Imagine being able to measure one of the most fundamental signs of human health—blood pressure—without a single cuff, without a wearable device digging into your skin. This sounds like science fiction, but in 2026, researchers are moving this possibility closer to reality. Measuring vital signs non-contact using radio waves, specifically with advanced radar systems, is a frontier of healthcare technology. Yet, turning the raw, noisy electronic signals from these radars into a precise, meaningful pulse waveform is an incredibly tricky engineering puzzle. It’s like trying to hear a faint whisper in a hurricane.

Turning Static Noise into Lifeline Signals: How Momentum Guides Radar to Perfect Blood Pressure Readings Imagine being able to measure one of the most fundamental signs of human health—blood pressure—without a single cuff, without a wearable device digging into your skin.

D3T: Dual-Domain Diffusion Transformer in Triplanar Latent Space for 3D Incomplete-View CT Reconstruction

Medical imaging, particularly Computed Tomography (CT), stands as a cornerstone of modern diagnostics. It allows doctors to peer inside the human body with remarkable detail, offering crucial insights into bone structures, implants, and organ health. However, the powerful imaging process of a standard CT scan often demands extensive radiation exposure and long acquisition times, making it impractical for quick checks, especially inside busy operating rooms. This is where the grand challenge of "incomplete-view" CT reconstruction comes into play—the science of rebuilding a complete 3D picture from only partial X-ray data.

Peering Into the Unseen: How New AI Brings Hidden Anatomy into Focus Medical imaging, particularly Computed Tomography (CT), stands as a cornerstone of modern diagnostics.

On the Utility of Foundation Models for Fast MRI: Vision-Language-Guided Image Reconstruction

In the relentless pursuit of better healthcare, the quality and speed of medical imaging are paramount. Magnetic Resonance Imaging (MRI) has long been a gold standard for visualizing the delicate structures of the human body, from the intricate pathways of the brain to the complex anatomy of the knee. However, traditional MRI scanning often demands long acquisition times. These long scans are not merely an inconvenience; they can lead to patient discomfort, motion artifacts, and, critically, limit the feasibility of routine screening, especially for mobile populations or those suffering from movement disorders. The dream, and the intense research focus, is simple: how can we acquire the necessary diagnostic detail in a fraction of the time?

Steering the Scanner: How AI is Guiding the Future of Fast MRI The Race Against Time in Medical Imaging

Robustness and Stability Analysis of Differentiable Shift-Variant FBP for Cone-Beam CT under Challenging Acquisition Settings

In the relentless pursuit of sharper, faster, and more precise medical diagnostics, the world of Computed Tomography (CT) is undergoing a breathtaking revolution. For decades, medical imaging relied on highly standardized procedures—think perfect, circular scans. These traditional methods are brilliant when everything is ideal, but the complexity of the human body, the needs of modern operating theaters, and the rise of sophisticated robotic scanners demand more flexibility. When metal implants are present, or when a scan path deviates from a perfect circle, old algorithms start to falter, producing blurry images and frustrating artifacts.

Steering Clear of Snags: Mastering CT Scans in the Wild In the relentless pursuit of sharper, faster, and more precise medical diagnostics, the world of Computed Tomography (CT) is undergoing a breathtaking revolution.

Convergent representations of linguistic constructions in human and artificial neural systems

For years, the dream of building true artificial intelligence has been intertwined with the enigma of the human mind. Can a machine truly grasp the nuance of language? Can our internal algorithms—whether chemical or silicon-based—arrive at the same core understanding of grammar? This question has driven cognitive science and machine learning into parallel, often competitive, arenas. Now, a remarkable 2026 study published in *Frontiers in Language Sciences* suggests that at a fundamental level concerning how we process language structure, the human brain and the most advanced artificial neural systems are coming to the same highly efficient answers. This research presents a compelling glimpse into how learning systems, regardless of their architecture, might naturally gravitate toward optimal solutions for understanding the building blocks of human expression.

When Brains and AI Meet: How Language's Hidden Grammar Converges in Thought and Code For years, the dream of building true artificial intelligence has been intertwined with the enigma of the human mind.