Rowan Cheung

Sub-mA Threshold III-V/Si Quantum Dot Microring Lasers Achieve Record Efficiency and Thermal Stability on Silicon

This work reports hybrid InAs/GaAs quantum dot microring lasers on silicon achieving sub-mA threshold currents (< 0.8 mA), record wall-plug efficiency (~10%), and an exceptionally high characteristic temperature (T0 = 212 K), demonstrating minimal thermal sensitivity. These results represent a significant advance in practical, energy-efficient on-chip laser sources for silicon photonic integrated circuits. The combination of ultra-low threshold, high efficiency, and temperature stability addresses key bottlenecks in deploying quantum dot lasers for dense wavelength-division multiplexing and high-speed (5 GHz bandwidth) interconnects.

Characterizing Detectability in 3DGS Poisoning: A Stage-wise Benchmark

AnchorMoE: Interpretable Time Series Classification via Anchor-Routed MoE

DyaPlex: Full-Duplex Speech-Motion Model for Dyadic Interaction

Thermal Positivity

First GeV Detection of a Superluminous Supernova Points to Magnetar Central Engine Over CSM Interaction

Fermi-LAT's systematic 16-year survey of nearby Type I and Type II superluminous supernovae yielded a single GeV detection: SN 2017egm, with >5σ significance (TS = 26–33). The gamma-ray signal — emerging 50–160 days post-explosion with a power-law spectrum (Γ = 2.17 ± 0.23) — is consistent with magnetar wind nebula models but cannot be explained by CSM interaction, which fails to reproduce both the signal timing and the observed Lγ/Lopt ~ 1 ratio (orders of magnitude above CSM-dominated objects). This constitutes the strongest observational evidence to date that a magnetar central engine drives the extreme luminosity of at least some SLSNe, and CTAO simulations suggest such events would be detectable out to 140 Mpc under the magnetar model.

18-dB on-chip vacuum squeezing in an adaptively poled lithium niobate waveguide

PEAM: Parametric Embodied Agent Memory through Contrastive Internalization of Experience in Minecraft

Can Segmentation Models Understand the World? Towards Proactive Affordance Reasoning via Visual Chain-of-Thought

A Non-Volatile Heterogeneous Quantum Dot III-V/Si DFB Laser with Optical Memristive Behavior

Pure Exploration for a Good Policy in Reinforcement Learning with Bandit Feedback

Tetris: Tile-level Sampling for Efficient and High-Fidelity Video Object Tracking

On Perturbatively Dressed Observables

Uncovering Intra-expert Activation Sparsity for Efficient Mixture-of-Expert Model Execution

Geometry Conflict: Explaining and Controlling Forgetting in LLM Continual Post-Training

Urban-ImageNet: A Large-Scale Multi-Modal Dataset and Evaluation Framework for Urban Space Perception

The Truth Lies Somewhere in the Middle (of the Generated Tokens)