Today's Tech News: Notable or Not?

https://www.eurekalert.org/news-releases/1129339 "software/ neuromorphic photonic hardware platform deploys optical devices such as Mach-Zehnder interferometer/ distributed feedback laser-saturable absorber chips, achieving rapid, light-speed parallel computing... optimizes routing configurations, reduces average packet delay, minimizes packet loss rate under high-load conditions... photonic spiking neural network/ machine learning integration offers promising alternative to traditional electronic routers for managing heavy data traffic"

https://techxplore.com/news/2026-04-compression-technique-ai-leaner-faster.html "shrinking AI expensive/ time-consuming: removing parts after training/ training small to copy big... overcome using state-space models identifying parts of model pulling their weight early in training/ removing dead weight... reverts to a previous checkpoint if performance drops... compressed models train up to 1.5X faster than full-sized... reduced to 25% original size 85.7% accurate, outperforming same small model trained from scratch... 40X faster and more accurate than other modern spectral techniques" Related: A hardware-software co-design can efficiently run AI on edge devices https://techxplore.com/news/2026-04-hardware-software-efficiently-ai-edge.html

https://techxplore.com/news/2026-04-method-neural-networks-faster-propagation.html "struggles with wave propagation because wave equations mathematically stiff, and standard AI models require many parameters/ data points, so extremely slow heavy, often requiring days of supercomputer time... overcome embedding phase/ amplitude physics into network's architecture... predicts wave variation rather than entire wave itself... coarse wave path model/ lets neural network fill in complex local interference patterns... energy exploration, medical Imaging, telecommunications" Related: AI-powered lab discovers brighter lead-free nanomaterials in 12 hours https://phys.org/news/2026-05-ai-powered-lab-brighter-free.html

https://techxplore.com/news/2026-04-memristor-built-oxygen-gradient-stability.html "memristor abrupt/ unstable state changes difficult for complex AI tasks like reinforcement learning... overcome with 2nd-order memristor using stable, built-in oxygen gradient using molecularly coordinated layer... creates dynamic barrier evolving very slowly (taking >100 seconds to shift) allowing information to process/ integrate longer... reduced required training iterations 68.75% (static environments)/ 35.65% (dynamic environments), -98.1% conductance modulation for highly granular synaptic weights' control"

https://phys.org/news/2025-12-quantum-machine-nears-partial-error.html "quantum processors noisy needing millions qubits to fix tiny errors accumulating in hundreds of gates required for AI models... overcome: don't need to correct every gate as more than half trainable, adapting to noise during learning... correcting only non-trainable gates achieves accuracy levels comparable to full error correction... requires only few thousand qubits instead of millions... quantum-powered AI capable of training on massive datasets in seconds could be achieved in years rather than decades"

https://www.eurekalert.org/news-releases/1109566 "logic (transistors)/ memory fabricated in 1 compact stack on 1 chip... 2 nm thick amorphous indium oxide active channel layer grown at relatively cool 150°C preventing damage to existing front-end circuitry, minimized defects (oxygen vacancies)... added ferroelectric hafnium-zirconium-oxide layer, ending up with 20 nm back-end transistors with built-in memory, 10 ns switching speed, lower voltage... generative AI, deep learning, computer vision... future: integrate these back-end memory transistors onto single circuit"

https://www.eurekalert.org/news-releases/1109250 "P-bits hardware elements naturally fluctuate between 0 & 1, so ideal for optimization/ inference, but rely on bulky, power-hungry, expensive digital-to-analog converters... overcome using magnetic tunnel junctions  naturally switching between 2 states randomly, a 50/50 bitstream fed into simple digital circuit that uses controlled timing combining signals, allowing smooth tuning probability of output being 0 or 1... requires significantly less area and power while being compatible with modern semiconductor manufacturing"

https://www.eurekalert.org/news-releases/1108851 "extreme thinness/ tendency to collapse due to metallic bonding makes synthesis difficult, overcome combining quantum-mechanical modeling/ machine learning... interface stability with smooth well-aligned geometries at metallene edges, most robust interfaces overall use transition metals, machine-learning accurately predicts atomic-level interface behavior... accelerates synthesis of more robust, larger-scale metallene structures... high-performance devices in: high-tech electronics, energy conversion, biomedicine"

https://techxplore.com/news/2025-11-intelligent-photodetectors-dogs-materials-spectra.html "directly encodes visible to mid-IR spectral/ spatial information into output photocurrent, so sensor identifies materials, chemicals,.. optoelectronic sensors learn/ compute without need for digital post-processing... electrically tunable enhancing/ suppressing specific spectral signatures... trains sampling pixels/ receiving simple commands, learns optimal electrical control sequence targeting object's pixels/ suppress background... for new/ unseen images, produces only for pixels belonging to target object" Related: Bioinspired robot eye adjusts its pupil to handle harsh lighting https://techxplore.com/news/2026-03-bioinspired-robot-eye-adjusts-pupil.html New chip lets robots see in 4d by tracking distance and speed simultaneously https://www.nature.com/articles/d41586-026-00727-1 Meta-optics enabling highly scalable machine vision https://www.eurekalert.org/news-releases/1120546...

https://www.eurekalert.org/news-releases/1093528 "circuit described using differential equations for how voltages/ currents evolve over time, taking into account real-world component characteristics... equations solved numerically, step by step, until steady-state... evaluates output voltage stability, power-delivery efficiency, total harmonic distortion, genetic algorithm then updates parameters, process repeated until optimum... <5% output voltage fluctuation (was 18%) when load changes... 86.7% power-delivery efficiency at 6.78 MHz, > 23 W output power"

https://spectrum.ieee.org/wireless-communication-over-air-processing "data collected from multiple devices at same time instead of from each separately, faster/ using fewer resources/ same wireless channel... signals naturally combine (forming an average) in the air, then picked up by receiver... interference among clusters alleviated combining 2 machine learning techniques: unfolding algorithm identifies optimized calculations for AI models, graph neural network identifies/ reduces signal interference across clusters... performed well handling 10 clusters... sensors, smart cities, sending updates" Related: New algorithm enables efficient machine learning with symmetric data structures https://techxplore.com/news/2025-07-algorithm-enables-efficient-machine-symmetric.html

https://www.eurekalert.org/news-releases/1084178 "combines neuro-inspired Hebbian machine learning computing with high-speed optoelectronics for high-speed / photonics adaptive feedback control based on input correlation learning rule... significantly higher frequencies enable low latency, high bandwidth, electromagnetic interference resistance... fiber-based dendritic structure with closed-loop controller, 1 GHz signaling/ sampling...  applied to a hypothetical temperature stabilization task, demonstrating its potential for ultra-fast, adaptive control” Related: Shape-shifting particles allow temperature control over fluid flow and stiffness https://phys.org/news/2025-07-shifting-particles-temperature-fluid-stiffness.html

https://source.washu.edu/2025/05/a-neuro-quantum-leap-in-finding-optimal-solutions/ "solves discovery problems (have new/ unknown solutions), the hardest in machine learning... Fowler-Nordheim annealers, using quantum mechanical tunneling principles setting search behavior (and when to shift where to look), guarantee to find optimal solution, and more reliability, given enough time... implemented on SpiNNaker2 neuromorphic computing platform... optimal protein folding/ configuration discovering new drugs, optimizing logistics in: supply chains, manufacturing, transportation services" Related: Neuromorphic devices and machine learning combine to make brain-like devices possible https://www.eurekalert.org/news-releases/1094358 Data storage: magnetic vortices in synthetic antiferromagnets move differently in 3D than in 2D https://www.eurekalert.org/news-releases/1100864 Lightweight design benchmark enables direct comparison of different methods https://techxplore.com/news/2025-1...

https://techxplore.com/news/2025-03-infomorphic-neurons-independently.html "neurons find their specific learning rules themselves, deciding which inputs are/ are not relevant to meet the scientist's very general, easy-to-understand learning goals... inspired by cerebral cortex's pyramidal cells... scientists focused on learning process of each individual neuron, applying information-theoretic measure precisely adjusting whether a neuron should seek more redundancy with its neighbors, collaborate synergistically, or try to specialize in its own part of the network's information... machine learning" Related: Artificial neurons become more advanced—and simpler—with conductive plastics https://techxplore.com/news/2025-09-artificial-neurons-advanced-simpler-plastics.html

https://www.eurekalert.org/news-releases/1074730 "conductive thermoplastics transmit electrical signals while providing structural support, but manufacture challenge controlling their internal structure due to bonding between filaments/ small cavities presence... 3D manufacture structures sensitive/ capable of transforming mechanical signals into electrical signals... enables use of softer materials, multifunctional material design... soft robotics, aerospace, biomedical/ infrastructure monitoring sensors that can obtain virtual data that can serve machine learning technologies" Related: New AI meA groundbreaking machine learning technique, Counterfactual https://www.eurekalert.org/news-releases/1080833

https://techxplore.com/news/2024-12-machine-tasks-retraining.html "algorithm generate output labels automatically from raw data... relies on pre-defined data augmentations, learning from general representation adapted to different transformations paying attention to context, representing abstract task/ environment... incorporates world models, invariant/ equivariant based on task at hand... transformer module encodes context, triplets representing previous experiences with transformations... natural language processing, computer vision, bioinformatics, speech recognition" Related: Researcher develops 'SpeechSSM,' opening up possibilities for a 24-hour AI voice assistant https://techxplore.com/news/2025-07-speechssm-possibilities-hour-ai-voice.html AI meets quantum: Machine learning advances estimation and control of quantum systems https://www.eurekalert.org/news-releases/1095181

https://www.eurekalert.org/news-releases/1071668 "sharp intersections/ corners cause stress concentrations/ early local failure/ breakage... multi-objective Bayesian optimization machine learning algorithm... 2-photon polymerization 3D printer optimized carbon nanolattices >2X strength, withstanding 2.03 MPa stress/ cubic mete per kg of its density, 5X> titanium... needed only 400 (not > 20k) data points... fuel savings of 80 liters per year for every kilogram of material you replace... low density/ high strength, stiffness... automotive, aerospace" Related: New Bayesian method enables rapid detection of quantum dot charge states https://www.eurekalert.org/news-releases/1082439

https://www.eurekalert.org/news-releases/1069246 "predicting properties such as electronic band gaps, formation energies, mechanical properties, needs training data which is limited because testing expensive time consuming... transfer learning predicts values: large model pre-trained on large dataset, then fine-tuned adapting to smaller target dataset, first learning simple task... performs much better than models trained from scratch... Multi-property Pre-Training on 7 different bulk 3D material properties... predict: band gap values, semiconductors forming point defects" Related: Machine learning algorithm enables faster, more accurate predictions on small tabular data sets https://techxplore.com/news/2025-01-machine-algorithm-enables-faster-accurate.html Tech companies are turning to 'synthetic data' to train AI models—but there's a hidden cost https://techxplore.com/news/2025-01-tech-companies-synthetic-ai-hidden.html Databricks Has a Trick That Lets AI Models...

https://phys.org/news/2024-12-chaotic-super-resolution-image-reconstruction.html   "traditional methods blur/ distort... Chaotic Mapping-based Sparse Representation overcomes noise/ computational complexity...  integrates into dictionary sequence, solving process of K-singular value decomposition dictionary update algorithm balancing traversal... simplifies search global optimal solutions, enhancing noise robustness... orthogonal matching pursuit greedy algorithm converges faster, complementing K-SVD, constructing high-resolution... reduced SR reconstruction complexity"

https://www.eurekalert.org/news-releases/1067967 "Single-Stream Image-to-Image Translation enables smartphone app... artistic styles, simulating weather changes, improving satellite video resolution, helping autonomous vehicles recognize different lighting... single encoder extracts spatial features, Direct Adaptive Instance Normalization with Pooling captures key style details focusing on the most prominent, improving efficiency... decoder takes combined content/ reconstructs final image... no expensive hardware or cloud services"