Chronological feed of everything captured from Manuel Endres.
paper / manuelendres / 9d ago
The paper introduces BIRCH-Trees, the first benchmark for individual tree height and species estimation from UAV RGB imagery, covering temperate, tropical, and boreal forest types. Alongside it, the authors present DINOvTree, a unified model leveraging a Vision Foundation Model (VFM) backbone with task-specific heads for joint height regression and species classification. DINOvTree achieves top overall performance while using only 54–58% of the parameters of the second-best competing approach, demonstrating that parameter efficiency and accuracy are not mutually exclusive in ecological remote sensing tasks.
computer-visionuav-remote-sensingforest-biomasstree-species-classificationvision-foundation-modelsenvironmental-monitoringdeep-learning
“BIRCH-Trees is the first benchmark dataset specifically designed for individual tree height and species estimation from tree-centered UAV imagery.”
paper / manuelendres / 9d ago
Agent4MR is an agentic framework that wraps general-purpose LLMs with structured, physics-aware validation to automatically generate and refine PyPulseq MRI pulse sequences. By embedding MRI physics constraints directly into the feedback loop, the system consistently produces artifact-free, physically valid sequences in a single user interaction — outperforming a context-only LLM baseline and matching or beating a human developer on interaction count. The framework also demonstrates autonomous "autoresearch" capability, where agents iteratively optimize sequences toward a target contrast without human intervention. This approach suggests that physics-grounded agentic harnesses may lower the barrier for non-experts to develop or innovate MR sequences.
llm-agentsmedical-imagingmri-sequence-designphysics-informed-aiai-automationmedical-physicscode-generation
“Agent4MR consistently produces artifact-free, physically valid MRI sequences in a single user interaction across all tested LLMs.”
paper / manuelendres / 9d ago
This paper presents an experimental radiation-hardness study of a spiking neural network (ODIN SNN processor) mapped onto flash-based FPGAs, exposed to a high-energy neutron beam to simulate space/avionics environments. The key finding is that enabling Spike-Dependent Synaptic Plasticity (SDSP) — an online learning mechanism — significantly delays application-level failure and allows partial self-recovery from accumulated single-event upsets (SEUs), compared to static inference-only configurations. The authors develop a calibrated fault model from measured SEU cross-sections and validate it via fault injection, providing a reusable methodology for radiation testing of neuromorphic processors with on-chip plasticity.
neuromorphic-computingspiking-neural-networksfpgaradiation-testingfault-tolerancehardware-architectureedge-ai
“Enabling SDSP (online learning) on the ODIN SNN processor significantly extends time to application-level failure under neutron radiation compared to inference-only mode.”
paper / manuelendres / 9d ago
E. coli chemotaxis has long been understood as operating near the physical limits of sensing (Berg-Purcell framework), but recent information-theoretic analyses reveal the bacterium uses only a small fraction of available ligand-arrival information. Endres argues this apparent inefficiency is not a performance failure — instead, the run-and-tumble movement strategy itself achieves robustness through symmetry and temporal averaging, decoupling behavioral performance from information processing efficiency. Comparing bacterial and eukaryotic chemotaxis further illustrates how different sensing architectures convert physical sensing limits into distinct behavioral outcomes, suggesting low information efficiency may reflect an evolved optimum balancing robustness, simplicity, and function.
bacterial-chemotaxisbiophysicsinformation-theorysignal-noisecell-behaviorsensing-limitsrun-and-tumble
“E. coli uses only a small fraction of the information available in ligand arrival statistics to bias its motion.”
paper / manuelendres / Mar 30
Shor's algorithm, crucial for cryptography, traditionally demands millions of physical qubits due to error correction overhead. However, new advancements in quantum error correction, logical instruction sets, and circuit design enable Shor's algorithm to be executed with as few as 10,000 reconfigurable atomic qubits. This reduction in qubit count makes cryptographically relevant quantum computation more feasible, with potential runtimes of a few days for P-256 elliptic curve discrete logarithms using 26,000 qubits.
quantum-computingshors-algorithmerror-correctionneutral-atom-qubitscryptography
“Shor's algorithm can be executed at cryptographically relevant scales with as few as 10,000 reconfigurable atomic qubits.”
paper / manuelendres / Mar 23
This work introduces an ancilla-based toolbox to address key hardware limitations in neutral atom quantum computing, specifically atom loss and heating. By leveraging Rydberg entangling gates and ancilla atoms, the authors demonstrate improved readout fidelity, coherence-preserving mid-circuit loss detection, and significant algorithmic cooling of data atoms. These advancements aim to enable more robust and continuous operation in optical tweezer experiments.
quantum-computingneutral-atomsrydberg-gatesquantum-error-correctionquantum-sensingatomic-physics
“The proposed ancilla-based toolbox mitigates atom loss and heating in optical tweezer experiments.”
paper / manuelendres / Jan 22
This work presents the first direct experimental observation of energy excitation spectra in emergent Conformal Field Theories (CFTs) at quantum phase transitions. The researchers developed and implemented a modulation technique to resolve the finite-size spectra of a Rydberg chain, which was tuned to quantum phase transitions described by Ising or tricritical Ising CFTs. This method allowed for the recovery of universal energy ratios characteristic of the underlying field theories, and provides a technique for diagnosing unknown universality classes in future experiments, advancing the understanding of CFT features in quantum simulators.
conformal-field-theoryquantum-physicsquantum-simulationrydberg-atomsphase-transitions
“The energy excitation spectra of emergent CFTs at quantum phase transitions can be directly observed experimentally.”
paper / manuelendres / Oct 13
Quantum simulations of high-energy gauge theory dynamics, specifically the (1+1)D Schwinger model, reveal surprising non-thermalizing behavior. Despite initial expectations of rapid thermalization for a system initialized with particle-antiparticle pairs, experiments show ballistic plasma formation and long-time memory effects. This necessitates a new theoretical framework involving plasma oscillations between electric field and current operators, departing from explanations like many-body scars.
quantum-simulationgauge-theoryhigh-energy-physicsplasma-physicsquantum-opticstheoretical-physics
“High-energy gauge theory dynamics can exhibit ballistic plasma formation and long-time memory effects.”
paper / manuelendres / Oct 9
Researchers have developed a three-dimensional acousto-optic deflector (AOD) by integrating a double-pass AOD with a diffraction grating in a Littrow configuration. This architecture creates a frequency-tunable lens capable of high-speed (100 kHz) axial and lateral beam steering and multiplexed multi-focal profile generation. The system effectively bridges the gap in rapid 3D beam shaping for applications in quantum technologies and high-resolution microscopy.
acousto-optic-deflectors3d-beam-shapingoptical-engineeringmicroscopylaser-processingquantum-technologies
“The system enables axial scanning over more than twenty Rayleigh ranges.”