One Fits Many: Class Confusion Loss for Versatile Domain Adaptation.

In the open world, various label sets and domain configurations give rise to a variety of Domain Adaptation (DA) setups, including closed-set, partial-set, open-set, and universal DA, as well as multi-source and multi-target DA. It is notable that existing DA methods are generally designed only for a specific setup, and may under-perform in setups they are not tailored to. This paper shifts the common paradigm of DA to Versatile Domain Adaptation (VDA), where one method can handle several different DA setups without any modification. Towards this goal, we first delve into a general inductive bias: class confusion, and then uncover that reducing such pairwise class confusion leads to significant transfer gains. With this insight, we propose one general class confusion loss (CC-Loss) to learn many setups. We estimate class confusion based only on classifier predictions and minimize the class confusion to enable accurate target predictions. Further, we improve the loss by enforcing the consistency of confusion matrices under different data augmentations to encourage its invariance to distribution perturbations. Experiments on 2D vision and 3D vision benchmarks show that the CC-Loss performs competitively in different mainstream DA setups. Code is available at https://github.com/thuml/Transfer-Learning-Library.

A new methodology for studying vortex dynamics based on point-contact spectroscopy.

Vortex dynamics has attracted tremendous attention for both fundamental physics and applications of type-II superconductors. However, methods to detect local vortex motion or vortex jump with high sensitivity are still scarce. Here, we fabricated soft point contacts on the clean layered superconductor 2H-NbSe2, which are demonstrated to contain multiple parallel micro-constrictions by scanning electronic microscopy. Andreev reflection spectroscopy was then studied in detail for the contacts. Differential conductance taken at fixed bias voltages was discovered to vary spontaneously over time in various magnetic fields perpendicular to the sample surface. The conductance variations become invisible when the field is zero or large enough, or parallel to the sample surface, which can be identified as the immediate consequence of vortex motion across a finite number of micro-constrictions. These results demonstrate point contact Andreev reflection spectroscopy to be a new potential way with a high time resolution to study the vortex dynamics in type-II superconductors.

A mixed-dimensional quasi-1D BiSeI nanowire-2D GaSe nanosheet p-n heterojunction for fast response optoelectronic devices.

Due to the unique combination configuration and the formation of a built-in electric field, mixed-dimensional heterojunctions present fruitful possibilities for improving the optoelectronic performances of low-dimensional optoelectronic devices. However, the response times of most photodetectors built from mixed-dimensional heterojunctions are within the millisecond range, limiting their applications in fast response optoelectronic devices. Herein, a mixed-dimensional BiSeI/GaSe van der Waals heterostructure is designed, which exhibits visible light detection ability and competitive photoresponsivity of 750 A W-1 and specific detectivity of 2.25 × 1012 Jones under 520 nm laser excitation. Excitingly, the device displays a very fast response time, e.g., the rise time and decay time under 520 nm laser excitation are 65 µs and 190 µs, respectively. Our findings provide a prospective approach to mixed-dimensional heterojunction photodetection devices with rapid switching capabilities.

A high-performance long-wave infrared photodetector based on a WSe2/PdSe2 broken-gap heterodiode.

Layered narrow bandgap quasi-two-dimensional (2D) transition metal dichalcogenides (TMDs) demonstrated excellent performance in long-wave infrared (LWIR) detection. However, the low light on/off ratio and specific detectivity (D*) due to the high dark current of the device fabricated using a single narrow bandgap material hindered its wide application. Herein, we report a type-III broken-gap band-alignment WSe2/PdSe2 van der Waals (vdW) heterostructure. The heterodiode device has a prominently low dark current and exhibits a high photoresponsivity (R) of 55.3 A W-1 and a high light on/off ratio >105 in the visible range. Notably, the WSe2/PdSe2 heterodiode shows an excellent uncooled LWIR response, with an R of ∼0.3 A W-1, a low noise equivalence power (NEP) of 4.5 × 10-11 W Hz-1/2, and a high D* of 1.8 × 108 cm Hz1/2 W-1. This work provides a new approach for designing high-performance room-temperature operational LWIR photodetectors.

Omni-Training: Bridging Pre-Training and Meta-Training for Few-Shot Learning.

Few-shot learning aims to fast adapt a deep model from a few examples. While pre-training and meta-training can create deep models powerful for few-shot generalization, we find that pre-training and meta-training focus respectively on cross-domain transferability and cross-task transferability, which restricts their data efficiency in the entangled settings of domain shift and task shift. We thus propose the Omni-Training framework to seamlessly bridge pre-training and meta-training for data-efficient few-shot learning. Our first contribution is a tri-flow Omni-Net architecture. Besides the joint representation flow, Omni-Net introduces two parallel flows for pre-training and meta-training, responsible for improving domain transferability and task transferability respectively. Omni-Net further coordinates the parallel flows by routing their representations via the joint-flow, enabling knowledge transfer across flows. Our second contribution is the Omni-Loss, which introduces a self-distillation strategy separately on the pre-training and meta-training objectives for boosting knowledge transfer throughout different training stages. Omni-Training is a general framework to accommodate many existing algorithms. Evaluations justify that our single framework consistently and clearly outperforms the individual state-of-the-art methods on both cross-task and cross-domain settings in a variety of classification, regression and reinforcement learning problems.

ModeRNN: Harnessing Spatiotemporal Mode Collapse in Unsupervised Predictive Learning.

Learning predictive models for unlabeled spatiotemporal data is challenging in part because visual dynamics can be highly entangled, especially in real scenes. In this paper, we refer to the multi-modal output distribution of predictive learning as spatiotemporal modes. We find an experimental phenomenon named spatiotemporal mode collapse (STMC) on most existing video prediction models, that is, features collapse into invalid representation subspaces due to the ambiguous understanding of mixed physical processes. We propose to quantify STMC and explore its solution for the first time in the context of unsupervised predictive learning. To this end, we present ModeRNN, a decoupling-aggregation framework that has a strong inductive bias of discovering the compositional structures of spatiotemporal modes between recurrent states. We first leverage a set of dynamic slots with independent parameters to extract individual building components of spatiotemporal modes. We then perform a weighted fusion of slot features to adaptively aggregate them into a unified hidden representation for recurrent updates. Through a series of experiments, we show high correlation between STMC and the fuzzy prediction results of future video frames. Besides, ModeRNN is shown to better mitigate STMC and achieve the state of the art on five video prediction datasets.

Skilful nowcasting of extreme precipitation with NowcastNet.

Extreme precipitation is a considerable contributor to meteorological disasters and there is a great need to mitigate its socioeconomic effects through skilful nowcasting that has high resolution, long lead times and local details1-3. Current methods are subject to blur, dissipation, intensity or location errors, with physics-based numerical methods struggling to capture pivotal chaotic dynamics such as convective initiation4 and data-driven learning methods failing to obey intrinsic physical laws such as advective conservation5. We present NowcastNet, a nonlinear nowcasting model for extreme precipitation that unifies physical-evolution schemes and conditional-learning methods into a neural-network framework with end-to-end forecast error optimization. On the basis of radar observations from the USA and China, our model produces physically plausible precipitation nowcasts with sharp multiscale patterns over regions of 2,048 km × 2,048 km and with lead times of up to 3 h. In a systematic evaluation by 62 professional meteorologists from across China, our model ranks first in 71% of cases against the leading methods. NowcastNet provides skilful forecasts at light-to-heavy rain rates, particularly for extreme-precipitation events accompanied by advective or convective processes that were previously considered intractable.

Enhanced Piezoelectricity and Thermal Stability of Electrostrain Performance in BiFeO3-Based Lead-Free Ceramics.

BiFeO3-based ceramics possess an advantage over large spontaneous polarization and high Curie temperature, and are thus widely explored in the field of high-temperature lead-free piezoelectrics and actuators. However, poor piezoelectricity/resistivity and thermal stability of electrostrain make them less competitive. To address this problem, (1 - x) (0.65BiFeO3-0.35BaTiO3)-xLa0.5Na0.5TiO3 (BF-BT-xLNT) systems are designed in this work. It is found that piezoelectricity is significantly improved with LNT addition, which is contributed by the phase boundary effect of rhombohedral and pseudocubic phase coexistence. The small-signal and large-signal piezoelectric coefficient (d33 and d33*) peaks at x = 0.02 with 97 pC/N and 303 pm/V, respectively. The relaxor property and resistivity are enhanced as well. This is verified by Rietveld refinement, dielectric/impedance spectroscopy and piezoelectric force microscopy (PFM) technique. Interestingly, a good thermal stability of electrostrain is obtained at x = 0.04 composition with fluctuation η = 31% (Smax'-SRTSRT×100%), in a wide temperature range of 25-180 °C, which is considered as a compromise of negative temperature dependent electrostrain for relaxors and the positive one for ferroelectric matrix. This work provides an implication for designing high-temperature piezoelectrics and stable electrostrain materials.

Ultrahigh-Sensitivity and Fast-Speed Solar-Blind Ultraviolet Photodetector Based on a Broken-Gap van der Waals Heterodiode.

Broad-bandgap semiconductor-based solar-blind ultraviolet (SBUV) photodetectors have attracted considerable research interest because of their broad applications in missile plume tracking, flame detectors, environmental monitoring, and optical communications due to their solar-blind nature and high sensitivity with low background radiation. Owing to its high light absorption coefficient, abundance, and wide tunable bandgap of 2-2.6 eV, tin disulfide (SnS2) has emerged as one of the most promising compounds for application in UV-visible optoelectronic devices. However, SnS2 UV detectors have some undesirable properties such as slow response speed, high current noise level, and low specific detectivity. This study reports a metal mirror-enhanced Ta0.01W0.99Se2/SnS2 (TWS) van der Waals heterodiode-based SBUV photodetector with an ultrahigh photoresponsivity (R) of ∼1.85 × 104 AW-1 and a fast speed with rising time (τr) of 3.3 µs and decay time (τd) of 3.4 µs. Notably, the TWS heterodiode device exhibits a significantly low noise equivalent power of ∼1.02 × 10-18 W Hz-1/2 and a high specific detectivity of ∼3.65 × 1014 cm Hz1/2 W-1. This study provides an alternative method for designing fast-speed SBUV photodetectors with enormous potential in applications.

From Big to Small: Adaptive Learning to Partial-Set Domains.

Domain adaptation targets at knowledge acquisition and dissemination from a labeled source domain to an unlabeled target domain under distribution shift. Still, the common requirement of identical class space shared across domains hinders applications of domain adaptation to partial-set domains. Recent advances show that deep pre-trained models of large scale endow rich knowledge to tackle diverse downstream tasks of small scale. Thus, there is a strong incentive to adapt models from large-scale domains to small-scale domains. This paper introduces Partial Domain Adaptation (PDA), a learning paradigm that relaxes the identical class space assumption to that the source class space subsumes the target class space. First, we present a theoretical analysis of partial domain adaptation, which uncovers the importance of estimating the transferable probability of each class and each instance across domains. Then, we propose Selective Adversarial Network (SAN and SAN++) with a bi-level selection strategy and an adversarial adaptation mechanism. The bi-level selection strategy up-weighs each class and each instance simultaneously for source supervised training, target self-training, and source-target adversarial adaptation through the transferable probability estimated alternately by the model. Experiments on standard partial-set datasets and more challenging tasks with superclasses show that SAN++ outperforms several domain adaptation methods.

PredRNN: A Recurrent Neural Network for Spatiotemporal Predictive Learning.

The predictive learning of spatiotemporal sequences aims to generate future images by learning from the historical context, where the visual dynamics are believed to have modular structures that can be learned with compositional subsystems. This paper models these structures by presenting PredRNN, a new recurrent network, in which a pair of memory cells are explicitly decoupled, operate in nearly independent transition manners, and finally form unified representations of the complex environment. Concretely, besides the original memory cell of LSTM, this network is featured by a zigzag memory flow that propagates in both bottom-up and top-down directions across all layers, enabling the learned visual dynamics at different levels of RNNs to communicate. It also leverages a memory decoupling loss to keep the memory cells from learning redundant features. We further propose a new curriculum learning strategy to force PredRNN to learn long-term dynamics from context frames, which can be generalized to most sequence-to-sequence models. We provide detailed ablation studies to verify the effectiveness of each component. Our approach is shown to obtain highly competitive results on five datasets for both action-free and action-conditioned predictive learning scenarios.

Polarity Modulation Induced High Electrostrain Performance with Near-Zero Hysteresis in a (Sr0.7Bi0.2□0.1)TiO3-Based System.

High-precision piezo actuators necessitate dielectrics with high electrostrain performance with low hysteresis. Polarity-modulated (Sr0.7Bi0.2□0.1)TiO3-based ceramics exhibit extraordinarily discrete multiphase coexistence regions: (i) the relaxor phase coexistence (RPC) region with local weakly polar tetragonal (T) and pseudocubic (Pc) short-range polar nanodomains and (ii) the ferroelectric phase coexistence (FPC) region with T long-range domains and Pc nanodomains. The RPC composition features a specially high and pure electrostrain performance with near-zero hysteresis (S ∼ 0.185%, Q33 ∼ 0.038 m4·C-2), which is double those of conventional Pb(Mg1/3Nb2/3)O3-based ceramics. Particular interest is paid to the RPC and FPC with multiscale characterization to unravel local structure-performance relationships. Guided by piezoelectric force microscopy, scanning transmission electron microscopy, and phase-field simulations, the RPC composition with multiphase low-angle weakly polar nanodomains shows local structural heterogeneity and contributes to a flat local free energy profile and thus to nanodomain switching and superior electrostrain performance, in contrast to the FPC composition with a macroscopic domain that shows stark hysteresis. This work provides a paradigm to design high-precision actuator materials with large electrostrain and ultralow hysteresis, extending our knowledge of multiphase coexistence species in ferroelectrics.

Fully Depleted Self-Aligned Heterosandwiched Van Der Waals Photodetectors.

Room-temperature-operating highly sensitive mid-wavelength infrared (MWIR) photodetectors are utilized in a large number of important applications, including night vision, communications, and optical radar. Many previous studies have demonstrated uncooled MWIR photodetectors using 2D narrow-bandgap semiconductors. To date, most of these works have utilized atomically thin flakes, simple van der Waals (vdW) heterostructures, or atomically thin p-n junctions as absorbers, which have difficulty in meeting the requirements for state-of-the-art MWIR photodetectors with a blackbody response. Here, a fully depleted self-aligned MoS2 -BP-MoS2 vdW heterostructure sandwiched between two electrodes is reported. This new type of photodetector exhibits competitive performance, including a high blackbody peak photoresponsivity up to 0.77 A W-1 and low noise-equivalent power of 2.0 × 10-14  W Hz-1/2 , in the MWIR region. A peak specific detectivity of 8.61 × 1010  cm Hz1/2  W-1 under blackbody radiation is achieved at room temperature in the MWIR region. Importantly, the effective detection range of the device is twice that of state-of-the-art MWIR photodetectors. Furthermore, the device presents an ultrafast response of ≈4 µs both in the visible and short-wavelength infrared bands. These results provide an ideal platform for realizing broadband and highly sensitive room-temperature MWIR photodetectors.

Protonation enhanced superconductivity in PdTe2.

Electrochemical ionic liquid gating is an effective way to intercalate ions into layered materials and modulate the properties. Here we report an enhanced superconductivity in a topological superconductor candidate PdTe2through electrochemical gating procedure. The superconducting transition temperature was increased to approximately 3.2 K by ionic gating induced protonation at room temperature. Moreover, a further enhanced superconductivity of both superconducting transition temperature and superconducting volume fraction was observed after the gated samples were placed in a glove box for 2 months. This may be caused by the diffusion of protons in the gated single crystals, which is rarely reported in electrochemical ionic liquid gating experiments. Our results further the superconducting study of PdTe2and may reveal a common phenomenon in the electrochemical gating procedure.

A Submicrosecond-Response Ultraviolet-Visible-Near-Infrared Broadband Photodetector Based on 2D Tellurosilicate InSiTe3.

2D material (2DM) based photodetectors with broadband photoresponse are of great value for a vast number of applications such as multiwavelength photodetection, imaging, and night vision. However, compared with traditional photodetectors based on bulk material, the relatively slow speed performance of 2DM based photodetectors hinders their practical applications. Herein, a submicrosecond-response photodetector based on ternary telluride InSiTe3 with trigonal symmetry and layered structure was demonstrated in this study. The InSiTe3 based photodetectors exhibit an ultrafast photoresponse (545-576 ns) and broadband detection capabilities from the ultraviolet (UV) to the near-infrared (NIR) optical communication region (365-1310 nm). Besides, the photodetector presents an outstanding reversible and stable photoresponse in which the response performance remains consistent within 200 000 cycles of switch operation. These significant findings suggest that InSiTe3 can be a promising candidate for constructing fast response broadband 2DM based optoelectronic devices.

VideoDG: Generalizing Temporal Relations in Videos to Novel Domains.

This paper introduces video domain generalization where most video classification networks degenerate due to the lack of exposure to the target domains of divergent distributions. We observe that the global temporal features are less generalizable, due to the temporal domain shift that videos from other unseen domains may have an unexpected absence or misalignment of the temporal relations. This finding has motivated us to solve video domain generalization by effectively learning the local-relation features of different timescales that are more generalizable, and exploiting them along with the global-relation features to maintain the discriminability. This paper presents the VideoDG framework with two technical contributions. The first is a new deep architecture named the Adversarial Pyramid Network, which improves the generalizability of video features by capturing the local-relation, global-relation, and cross-relation features progressively. On the basis of pyramid features, the second contribution is a new and robust approach of adversarial data augmentation that can bridge different video domains by improving the diversity and quality of augmented data. We construct three video domain generalization benchmarks in which domains are divided according to different datasets, different consequences of actions, or different camera views, respectively. VideoDG consistently outperforms the combinations of previous video classification models and existing domain generalization methods on all benchmarks.

Reversible engineering of topological insulator surface state conductivity through optical excitation.

Despite the broadband response, limited optical absorption at a particular wavelength hinders the development of optoelectronics based on Dirac fermions. Heterostructures of graphene and various semiconductors have been explored for this purpose, while non-ideal interfaces often limit the performance. The topological insulator (TI) is a natural hybrid system, with the surface states hosting high-mobility Dirac fermions and the small-bandgap semiconducting bulk state strongly absorbing light. In this work, we show a large photocurrent response from a field effect transistor device based on intrinsic TI Sn-Bi1.1Sb0.9Te2S (Sn-BSTS). The photocurrent response is non-volatile and sensitively depends on the initial Fermi energy of the surface state, and it can be erased by controlling the gate voltage. Our observations can be explained with a remote photo-doping mechanism, in which the light excites the defects in the bulk and frees the localized carriers to the surface state. This photodoping modulates the surface state conductivity without compromising the mobility, and it also significantly modify the quantum Hall effect of the surface state. Our work thus illustrates a route to reversibly manipulate the surface states through optical excitation, shedding light into utilizing topological surface states for quantum optoelectronics.

Air-Stable Low-Symmetry Narrow-Bandgap 2D Sulfide Niobium for Polarization Photodetection.

Low-symmetry 2D materials with unique anisotropic optical and optoelectronic characteristics have attracted a lot of interest in fundamental research and manufacturing of novel optoelectronic devices. Exploring new and low-symmetry narrow-bandgap 2D materials will be rewarding for the development of nanoelectronics and nano-optoelectronics. Herein, sulfide niobium (NbS3 ), a novel transition metal trichalcogenide semiconductor with low-symmetry structure, is introduced into a narrowband 2D material with strong anisotropic physical properties both experimentally and theoretically. The indirect bandgap of NbS3 with highly anisotropic band structures slowly decreases from 0.42 eV (monolayer) to 0.26 eV (bulk). Moreover, NbS3 Schottky photodetectors have excellent photoelectric performance, which enables fast photoresponse (11.6 µs), low specific noise current (4.6 × 10-25 A2 Hz-1 ), photoelectrical dichroic ratio (1.84) and high-quality reflective polarization imaging (637 nm and 830 nm). A room-temperature specific detectivity exceeding 107 Jones can be obtained at the wavelength of 3 µm. These excellent unique characteristics will make low-symmetry narrow-bandgap 2D materials become highly competitive candidates for future anisotropic optical investigations and mid-infrared optoelectronic applications.

Ultrafast and Sensitive Self-Powered Photodetector Featuring Self-Limited Depletion Region and Fully Depleted Channel with van der Waals Contacts.

Self-powered photodetectors with great potential for implanted medical diagnosis and smart communications have been severely hindered by the difficulty of simultaneously achieving high sensitivity and fast response speed. Here, we report an ultrafast and highly sensitive self-powered photodetector based on two-dimensional (2D) InSe, which is achieved by applying a device architecture design and generating ideal Schottky or ohmic contacts on 2D layered semiconductors, which are difficult to realize in the conventional semiconductors owing to their surface Fermi-level pinning. The as-fabricated InSe photodiode features a maximal lateral self-limited depletion region and a vertical fully depleted channel. It exhibits a high detectivity of 1.26 × 1013 Jones and an ultrafast response speed of ∼200 ns, which breaks the response speed limit of reported self-powered photodetectors based on 2D semiconductors. The high sensitivity is achieved by an ultralow dark current noise generated from the robust van der Waals (vdW) Schottky junction and a high photoresponsivity due to the formation of a maximal lateral self-limited depletion region. The ultrafast response time is dominated by the fast carrier drift driven by a strong built-in electric field in the vertical fully depleted channel. This device architecture can help us to design high-performance photodetectors utilizing vdW layered semiconductors.

Graphene Hybrid Structures for Integrated and Flexible Optoelectronics.

Graphene (Gr) has many unique properties including gapless band structure, ultrafast carrier dynamics, high carrier mobility, and flexibility, making it appealing for ultrafast, broadband, and flexible optoelectronics. To overcome its intrinsic limit of low absorption, hybrid structures are exploited to improve the device performance. Particularly, van der Waals heterostructures with different photosensitive materials and photonic structures are very effective for improving photodetection and modulation efficiency. With such hybrid structures, Gr hybrid photodetectors can operate from ultraviolet to terahertz, with significantly improved R (up to 109 A W-1 ) and bandwidth (up to 128 GHz). Furthermore, integration of Gr with silicon (Si) complementary metal-oxide-semiconductor (CMOS) circuits, the human body, and soft tissues is successfully demonstrated, opening promising opportunities for wearable sensors and biomedical electronics. Here, the recent progress in using Gr hybrid structures toward high-performance photodetectors and integrated optoelectronic applications is reviewed.