Chronological feed of everything captured from Jian-Wei Pan.
Researchers developed a specialized vacuum apparatus with compact 2D-MOT sources, modular science chamber, in-vacuum electrodes, and high-resolution imaging to prepare a quasi-2D degenerate mixture of ultracold 23Na and 87Rb atoms. The mixture is loaded into a single layer of a vertical optical lattice from a dual-species condensate, enabling observation of quantum immiscibility in equilibrium density profiles. These profiles match mean-field theory predictions, providing a platform for studying quantum impurities, droplets, and polar molecules in reduced dimensions.
SEA-Eval introduces the first benchmark for self-evolving agents (SEA), evaluating intra-task reliability and long-term evolution via sequential task streams measuring success rate and token consumption. Current LLM agents suffer from episodic amnesia and static tools, limiting cross-task learning. Empirical results show state-of-the-art frameworks exhibit up to 31.2x token consumption differences despite identical success rates, revealing divergent evolutionary trajectories.
TriDeliver introduces a hierarchical framework integrating human couriers, UAVs, and crowdsourced ground vehicles to overcome individual agent limitations in instant delivery. It employs transfer learning to distill courier behavioral knowledge into initial scheduling policies for UAVs and GVs, refined via fine-tuning for cooperative dispatching. Real-world evaluations show 65.8% cost reduction over UAV-courier baselines and further gains in time, cost, and GV task disruption using simple neural representations.
FDSM addresses spectral bias in diffusion models for zero-shot skeleton action recognition (ZSAR) by integrating a Semantic-Guided Spectral Residual Module, Timestep-Adaptive Spectral Loss, and Curriculum-based Semantic Abstraction. These components recover high-frequency motion details lost due to oversmoothing, enabling precise skeleton-text matching without labeled data. The method achieves state-of-the-art results on NTU RGB+D, PKU-MMD, and Kinetics-skeleton benchmarks.
Immersive Volumetric Videos (IVV) is a new format delivering high-res, high-frame-rate 6-DoF visual-auditory VR from real-world multi-view captures. The ImViD dataset provides 5K 60FPS videos (1-5 min) from a custom synchronized multi-modal rig, capturing complex dynamic indoor/outdoor scenes with rich interactions. A Gaussian-based dynamic light field reconstructor uses flow-guided init, temporal calibration, and spatio-temporal supervision; paired with novel sound field reconstruction for end-to-end IVV production. Benchmarks confirm superior quality, stability, and interaction space over priors.
This paper presents a novel approach to Twin-Field Quantum Key Distribution (TF-QKD) that overcomes the scalability limitations of wavelength-division multiplexing (WDM) by utilizing independent dissipative Kerr soliton (DKS) microcombs. This method simplifies multi-wavelength sources by requiring only stabilization of pump wavelength and repetition rates, leading to a significant enhancement in secure key rates over long distances.
Researchers have developed a high-performance quantum frequency conversion (QFC) system that bridges the spectral gap between ultraviolet (UV) photons and the telecom C-band. This system, based on thin-film lithium niobate, exhibits a record-high external efficiency of 28.8% and ultra-low noise of 35 counts per second. This advancement is critical for enabling long-lived remote ion-ion entanglement in scalable quantum networks by allowing for efficient information transfer between quantum systems operating at different wavelengths.
This paper introduces a neutral-atom quantum computing framework that addresses the challenges of cumulative motional heating and atom loss in deep quantum circuits. By integrating mid-circuit operations such as Raman sideband cooling and qubit re-initialization, the framework maintains high gate fidelities (~99.8%) across multiple operational rounds. This active management of internal and motional entropy provides a crucial pathway for scalable quantum error correction.
Researchers have successfully demonstrated device-independent quantum key distribution (DI-QKD) over a 100 km fiber optic link using single atoms. This was achieved by employing single-photon interference for entanglement heralding and quantum frequency conversion to mitigate fiber loss. The experiment also utilized a Rydberg-based emission scheme to suppress photon recoil, resulting in high-fidelity atom-atom entanglement and positive asymptotic key rates.
Entanglement distribution in quantum networks is bottlenecked by rapid decoherence in remote memory-memory entanglement. This research addresses this by developing long-lived trapped-ion memories, an efficient telecom interface, and a high-visibility single-photon entanglement protocol. This approach enables the establishment and maintenance of memory-memory entanglement over 10 km, demonstrating viable metropolitan-scale device-independent quantum key distribution.