Inverse-designed ultra-compact multi-channel and multi-mode waveguide crossings.

In this work, we use the inverse design method to design three-channel and four-channel dual-mode waveguide crossings with the design regions of 4.32 µm-wide regular hexagon and 6.68 µm-wide regular octagon, respectively. Based on the highly-symmetric structures, the fundamental transverse electric (TE0) and TE1 modes propagate through the waveguide crossings efficiently. Moreover, the devices are practically fabricated and experimentally characterized. The measured insertion losses and crosstalks of the three-channel and dual-mode waveguide crossing for both the TE0 and TE1 modes are less than 1.8 dB and lower than -18.4 dB from 1540 nm to 1560 nm, respectively. The measured insertion losses of the four-channel and dual-mode waveguide crossing for the TE0 and TE1 modes are less than 1.8 dB and 2.5 dB from 1540 nm to 1560 nm, respectively, and the measured crosstalks are lower than -17.0 dB. In principle, our proposed scheme can be extended to waveguide crossing with more channels and modes.

Three-dimensional mode-division multiplexing system.

Blindly increasing the channels of the mode (de)multiplexer on the single-layer chip can cause the device structure to be too complex to optimize. The three-dimensional (3D) mode division multiplexing (MDM) technology is a potential solution to extend the data capacity of the photonic integrated circuit by assembling the simple devices in the 3D space. In our work, we propose a 16 × 16 3D MDM system with a compact footprint of about 100 µm × 5.0 µm × 3.7 µm. It can realize 256 mode routes by converting the fundamental transverse electric (TE0) modes in arbitrary input waveguides into the expected modes in arbitrary output waveguides. To illustrate its mode-routing principle, the TE0 mode is launched in one of the sixteen input waveguides, and converted into corresponding modes in four output waveguides. The simulated results indicate that the ILs and CTs of the 16 × 16 3D MDM system are less than 3.5 dB and lower than -14.2 dB at 1550 nm, respectively. In principle, the 3D design architecture can be scaled to realize arbitrary network complexity levels.

Different-mode power splitters based on a multi-dimension direct-binary-search algorithm.

In this work, we design, fabricate, and characterize a different-mode (waveguide-connected) power splitter ((W)PS) by what we believe to be a novel multi-dimension direct-binary-search algorithm that can significantly balance the device performance, time cost, and fabrication robustness by searching the state-dimension, rotation-dimension, shape-dimension, and size-dimension parameters. The (W)PS can simultaneously generate the fundamental transverse electric (TE0) and TE1 mode with the 1:1 output balance. Compared with the PS, the WPS can greatly shorten the adiabatic taper length between the single-mode waveguide and the grating coupler. The measured results of the different-mode (W)PS indicate that the insertion loss and crosstalk are less than 0.9 (1.3) dB and lower than -17.8 (-14.9) dB from 1540 nm to 1560 nm. In addition, based on the tunable tap couplers, the different-mode (W)PS can be extended to multiple output ports with different modes and different transmittances.

Deep-learning-assisted linearization for the broadband photonic scanning channelized receiver.

An autoencoder-residual (AE-Res) network is designated to assist the linearization of the wideband photonic scanning channelized receiver. It is capable of adaptively suppressing spurious distortions over multiple octaves of signal bandwidth, obviating the need for calculating the multifactorial nonlinear transfer functions. Proof-of-concept experiments indicate that the improvement of the third-order spur-free dynamic range (SFDR2/3) is 17.44 dB. Moreover, the results for real wireless communication signals demonstrate that the improvement of the spurious suppression ratio (SSR) is 39.69 dB and the reduction of the noise floor is ∼10 dB.

CAEVT: Convolutional Autoencoder Meets Lightweight Vision Transformer for Hyperspectral Image Classification.

Convolutional neural networks (CNNs) have been prominent in most hyperspectral image (HSI) processing applications due to their advantages in extracting local information. Despite their success, the locality of the convolutional layers within CNNs results in heavyweight models and time-consuming defects. In this study, inspired by the excellent performance of transformers that are used for long-range representation learning in computer vision tasks, we built a lightweight vision transformer for HSI classification that can extract local and global information simultaneously, thereby facilitating accurate classification. Moreover, as traditional dimensionality reduction methods are limited in their linear representation ability, a three-dimensional convolutional autoencoder was adopted to capture the nonlinear characteristics between spectral bands. Based on the aforementioned three-dimensional convolutional autoencoder and lightweight vision transformer, we designed an HSI classification network, namely the "convolutional autoencoder meets lightweight vision transformer" (CAEVT). Finally, we validated the performance of the proposed CAEVT network using four widely used hyperspectral datasets. Our approach showed superiority, especially in the absence of sufficient labeled samples, which demonstrates the effectiveness and efficiency of the CAEVT network.

A Lightweight 1-D Convolution Augmented Transformer with Metric Learning for Hyperspectral Image Classification.

Hyperspectral image (HSI) classification is the subject of intense research in remote sensing. The tremendous success of deep learning in computer vision has recently sparked the interest in applying deep learning in hyperspectral image classification. However, most deep learning methods for hyperspectral image classification are based on convolutional neural networks (CNN). Those methods require heavy GPU memory resources and run time. Recently, another deep learning model, the transformer, has been applied for image recognition, and the study result demonstrates the great potential of the transformer network for computer vision tasks. In this paper, we propose a model for hyperspectral image classification based on the transformer, which is widely used in natural language processing. Besides, we believe we are the first to combine the metric learning and the transformer model in hyperspectral image classification. Moreover, to improve the model classification performance when the available training samples are limited, we use the 1-D convolution and Mish activation function. The experimental results on three widely used hyperspectral image data sets demonstrate the proposed model's advantages in accuracy, GPU memory cost, and running time.

RoMAT: Role-based multi-agent transformer for generalizable heterogeneous cooperation.

Multi-task multi-agent systems (MASs) are challenging to model because they involve heterogeneous agents with different behavior patterns that need to cooperate across various tasks. Existing networks for single-agent policies are not suitable for this setting, as they cannot share policies among agents without losing task-specific performance. We propose a novel framework called Role-based Multi-Agent Transformer (RoMAT), which uses a sequence modeling technique and a role-based actor to enable agents to adapt to different tasks and roles in MASs. RoMAT has a modular model architecture, where backbone networks are shared by all agents, but a small part of the parameters (role-based actor) is independent, depending on the agents' exclusive structures. We evaluate RoMAT on several benchmark tasks and show that it can capture the behavior patterns of heterogeneous agents and achieve better performance and generalization than other methods in both single and multi-task settings.

Psychiatric disorders associated with immune checkpoint inhibitors: a pharmacovigilance analysis of the FDA Adverse Event Reporting System (FAERS) database.

Background: With the increasing use of immune checkpoint inhibitors (ICIs) for tumour immunotherapy, the immune-related adverse events (irAEs) caused by their collateral effect on the immune system pose a key challenge for the clinical application of ICIs. Psychiatric adverse events are a class of adverse events associated with ICIs that are realistically observed in the real world. We aim to provide a comprehensive study and summary of psychiatric adverse events associated with ICIs. Methods: We obtained ICI adverse reaction reports during January 2012-December 2021 from the FDA Adverse Event Reporting System (FAERS) database. ICI reports underwent screening to minimize the influence of other adverse reactions, concomitant medications, and indications for medication use that may also contribute to psychiatric disorders. Disproportionality analysis was performed to find psychiatric adverse events associated with ICIs by comparing ICIs with the full FAERS database using the reporting odds ratio (ROR). Influencing factors were explored based on univariate logistic regression analysis. Finally, the Cancer Genome Atlas (TCGA) pan-cancer transcriptome data were combined to explore the potential biological mechanisms associated with ICI-related pAEs. Findings: Reports of psychiatric adverse events accounted for 2.71% of the overall ICI adverse event reports in the FAERS database. Five categories of psychiatric adverse events were defined as ICI-related psychiatric adverse events (pAEs). The median age of reports with ICI-related pAEs was 70 (interquartile range [IQR] 24-95), with 21.54% of reports having a fatal outcome. Cases with indications for lung cancer, skin cancer and kidney site cancer accounted for the majority. The odds of ICI-related pAEs increased in older patients (65-74: OR = 1.44 [1.22-1.70], P < 0.0001: ≥75: OR = 1.84 [1.54-2.20], P < 0.0001). The occurrence of ICI-related pAEs may be related to NOTCH signalling and dysregulation of synapse-associated pathways. Interpretation: This study investigated psychiatric adverse events highly associated with ICI treatment, their influencing factors and potential biological mechanisms, which provides a reliable basis for further in-depth study of ICI-related pAEs. However, as an exploratory study, our findings need to be further confirmed in a large-scale prospective study. Funding: This work was supported by the Natural Science Foundation of Guangdong Province (2018A030313846 and 2021A1515012593), the Science and Technology Planning Project of Guangdong Province (2019A030317020) and the National Natural Science Foundation of China (81802257, 81871859, 81772457, 82172750 and 82172811). Guangdong Basic and Applied Basic Research Foundation (Guangdong - Guangzhou Joint Fouds) (2022A1515111212). This work was supported by Key Research and Development Projects of Sichuan Science and Technology (2022YFS0221, 2022YFS0074, 2022YFS0156 and 2022YFS0378). Sichuan Provincial People's Hospital Hospital Young Talent Fund (2021QN08).

Tunable Infrared Emissivity in Multilayer Graphene by Ionic Liquid Intercalation.

Controllably tuned infrared emissivity has attracted great interest for potential application in adaptive thermal camouflage. In this work, we report a flexible multilayer graphene based infrared device on a porous polyethylene membrane, where the infrared emissivity could be tuned by ionic liquid intercalation. The Fermi level of surface multilayer graphene shifts to a high energy level through ionic liquid intercalation, which blocks electronic transition below the Fermi level. Thus, the optical absorptivity/emissivity of graphene could be controlled by intercalation. Experimentally, the infrared emissivity of surface graphene was found to be tuned from 0.57 to 0.41 after ionic liquid intercalation. Meanwhile, the relative reflectivity Rv/R0 of surface graphene increased from 1.0 to 1.15. The strong fluorescence background of Raman spectra, the upshift of the G peak (~23 cm-1), and the decrease of sheet resistance confirmed the successful intercalation of ionic liquid into the graphene layers. This intercalation control of the infrared emissivity of graphene in this work displays a new way of building an effective thermal camouflage system.