Search

Interactive robots that cooperate with humans must take appropriate actions in response to their requests. Unfortunately, such requests often have information gaps with their actual demands. However, robots are still expected to reason and act on what is required, depending on the situation. We call these reflective actions. To achieve such reflective actions for robots, we constructed a dataset that consists of the reflective actions of a domestic manipulation robot, in which the actions correspond to user utterances with their surroundings situations. By crowdsourcing, we defined several action scenarios that could be regarded as reflective. We recorded videos of situations described in the crowdsourcing scenarios, corresponding to the user situations just before the robot’s reflective actions. We also annotated the videos of the user utterance transcriptions, objects, user poses, and user positions to investigate the contribution of such descriptive features to the reflective action decisions. Our experimental results indicated that even though our newly defined task is very challenging, it can be solved if the system has a concrete understanding of the situation.

In explainable artificial intelligence (XAI) for object detection, saliency maps are employed to highlight important regions for a learned model’s prediction. However, a trade-off exists wherein the higher the accuracy of explanation results, the higher the computational cost, posing a challenge for practical applications. Therefore, this study proposes a novel XAI method for object detection to address this challenge. In recent years, research on XAI that satisfies desirable properties for explanatory validity by introducing the Shapley value has been widely conducted. However, a common drawback across these approaches is the high computational cost, which has hindered broad implementation. Our proposed method utilizes an explainer model that learns to estimate the Shapley value and provides a reliable explanation for object detection in a real-time inference. This framework can be applied to various object detectors in a model-agnostic manner. Through quantitative evaluation, we experimentally demonstrate that our method achieves the fastest explanation while delivering superior performance compared with other existing methods.

Ising machines are specialized computers for finding the lowest energy states of Ising spin models, onto which many practical combinatorial optimization problems can be mapped. Simulated bifurcation (SB) is a quantum-inspired parallelizable algorithm for Ising problems that enables scalable multi-chip implementations of Ising machines. However, the computational performance of a previously proposed multi-chip architecture tends to saturate as the number of chips increases for a given problem size because computation and communication are exclusive in the time domain. In this paper, we propose a streaming architecture for multi-chip implementations of SB-based Ising machines with full spin-to-spin connectivity. The data flow in in-chip computation is harmonized with the data flow in inter-chip communication, enabling the computation and communication to overlap and the communication time to be hidden. Systematic experiments demonstrate linear strong scaling of performance up to the vicinity of the ideal communication limit determined only by the latency of chip-to-chip communication. Our eight-FPGA (field-programmable gate array) cluster can compute a 32,768-spin problem with a high pipeline efficiency of 97.9%. The performance of a 79-FPGA cluster for a 100,000-spin problem, projected using a theoretical performance model validated on smaller experimental clusters, is comparable to that of a state-of-the-art 100,000-spin optical Ising machine.

Fine-grained fashion retrieval searches for items that share a similar attribute with the query image. Most existing methods use a pre-trained feature extractor (e.g., ResNet 50) to capture image representations. However, a pre-trained feature backbone is typically trained for image classification and object detection, which are fundamentally different tasks from fine-grained fashion retrieval. Therefore, existing methods suffer from a feature gap problem when directly using the pre-trained backbone for fine-tuning. To solve this problem, we introduce an attribute-guided multi-level attention network (AG-MAN). Specifically, we first enhance the pre-trained feature extractor to capture multi-level image embedding, thereby enriching the low-level features within these representations. Then, we propose a classification scheme where images with the same attribute, albeit with different values, are categorized into the same class. This can further alleviate the feature gap problem by perturbing object-centric feature learning. Moreover, we propose an improved attribute-guided attention module for extracting more accurate attribute-specific representations. Our model consistently outperforms existing attention based methods when assessed on the FashionAI (62.8788% in MAP), DeepFashion (8.9804% in MAP), and Zappos50k datasets (93.32% in Prediction accuracy). Especially, ours improves the most typical ASENet_V2 model by 2.12%, 0.31%, and 0.78% points in FashionAI, DeepFashion, and Zappos50k datasets, respectively. The source code is available in https://github.com/Dr-LingXiao/AG-MAN.

Explainable artificial intelligence (XAI) has witnessed significant advances in the field of object recognition, with saliency maps being used to highlight image features relevant to the predictions of learned models. Although these advances have made artificial intelligence (AI)-based technology more interpretable to humans, several issues have come to light, as some approaches present explanations irrelevant to predictions, and cannot guarantee the validity of XAI (axioms). In this study, we propose the Baseline Shapley-based Explainable Detector (BSED), which extends the Shapley value to object detection for images, thereby enhancing the validity of interpretation. The Shapley value can attribute the prediction of a learned model to a baseline feature while satisfying the explainability axioms. The processing cost for the BSED is within the reasonable range, while the original Shapley value is prohibitively computationally expensive. Furthermore, BSED is a generalizable method that can be applied to various object detectors for images in a model-agnostic manner, and interpret various detection targets without fine-grained parameter tuning. These strengths can enable the practical applicability of XAI. We present quantitative and qualitative evaluations to demonstrate that our method outperforms existing methods in terms of explanation validity. Moreover, we present some applications, such as correcting detection based on explanations from our method.

Recently, visible light communications (VLCs) have attracted attention as a complement to wireless communications by radio frequency. In addition, the demand for Internet of Things (IoT) communications has increased in the last decade, and VLC-based IoT systems are being considered. Because IoT devices operate at various data rates, multirate transmission is required. Herein, we propose a scheme that supports multirate data transmission for synchronous optical code division multiple access systems. This scheme employs a generalized modified prime sequence code and multiple extended bi-orthogonal codes. The proposed scheme realizes multirate transmission with an arbitrary number of data rate levels by superimposing spreading codes of different codelengths. In addition, the proposed scheme eliminates not only interference between users with the same data rate, but also interference between users with different data rates. Furthermore, we investigated the bit error rate performance and energy efficiency of the proposed scheme, considering the effects of background light, dark current, and PIN photodiode noise including shot noise, and thermal noise.

Developing dedicated hardware for each application in future optical access network systems causes significant initial investment costs. Thus, these costs are alleviated by optical access networks using general-purpose devices related to software-defined networking (SDN) and network function virtualization (NFV). We have studied physical layer virtualization to bring the advantages of SDN and NFV in optical access networks. We have focused on digital coherent technology to enhance capacity and extend reach within optical access networks. However, the future virtualized coherent access faces stringent frequency characteristics and the impact of chromatic dispersion (CD). The number of taps of finite impulse response (FIR) filters in the receiver increases rapidly for the CD compensation. The previously proposed DSP algorithm may not operate in this situation because the processing time exceeds the requirement. Therefore, this paper proposes and implements software for a lightweight multi-input-to-multi-output (MIMO) adaptive equalizer that separates FIR filters depending on the tap update period. The proposed method uses two types of FIR filters; one is used for polarization demultiplexing with short taps, and the other is a filter with long taps for the compensation of the CD in the fiber channel and the frequency distortion from the transceiver. Owing to filter separation, the calculation cost can be reduced. We indicate the processing time meeting desired requirements through real-time transmission experiments, even with increased taps. Additionally, we conduct offline simulations and experiments at a higher symbol rate than the real-time experiment.

Unlike traditional user-item recommendation tasks (e.g., movie or consumer-product recommendation), reciprocal recommender systems (RRSs) (e.g., online dating services and job-recruitment sites) must consider the interests of both two users. Pair matching prediction can improve the efficiency with which RRSs match potential partners. Graph Neural Networks (GNNs) are powerful models for learning representations of attributed graphs and information circulation between nodes. GNNs greatly facilitate link prediction in the area of user-item recommender systems but have not been extensively applied to RRS. In this study, we present a novel method for pair matching prediction that learns the reciprocal information circulation between users: not only side information about them but also structural information about their behavior histories. In contrast to earlier RRSs, which focus on response prediction, ours predicts both send and reply signals. Moreover, we introduce negative sample mining to explore the effect of different types of multiple samples on recommendation accuracy in real applications. Testing our method on data provided by an online dating service, we achieved an AUC of 73.15% (an absolute improvement of over 3.20% point above baseline) and an AP of 26.01% (an absolute improvement of over 2.79%) on send prediction; an AUC of 68.95% (an absolute improvement of over 1.74%) and an AP of 23.02% (an absolute improvement of over 0.70%) on reply prediction; an AUC of 71.26% (over 4.35% point absolute improvement) and an AP of 23.95% (over 0.30% point absolute improvement) on fusion reciprocal prediction.

Adverse events after surgery not only affect the patient’s recovery but also increase the burden on doctors and patients due to prolonged hospitalization. Predicting adverse events from patient data before surgery with a machine learning method is highly expected. It is difficult to collect a large amount of patient data since the number of surgeries in a year is limited and predict the occurrence of adverse events accurately since patient data are imbalanced data. To improve the accuracy of adverse event prediction, this paper proposes data generation with Filtered $\beta $ -VAE for the preoperative prediction of adverse events. Filtered $\beta $ -VAE has filters by the reconstruction error and by a machine learning method. After $\beta $ -VAE generates minority class data, the two layers of filtering are used to remove low-quality minority class data that have little contribution to the adverse event prediction. In the evaluations, patient data obtained from Tokyo Dental University Ichikawa General Hospital were used. The proposed method can predict adverse events with a recall of 0.848, which is 5.6% more accurate than existing methods. The effects of filtering in Filtered $\beta $ -VAE are visualized, and the reasons for the improvement in prediction accuracy are clarified. Furthermore, this paper shows that $\beta $ -VAE can generate arbitrary patient data even in table data, corresponding to the distribution of the original patient data.

Pairs-trading is a trading strategy that involves matching a long position with a short position in two stocks aiming at market-neutral profits. While a typical pairs-trading system monitors the prices of two statistically correlated stocks for detecting a temporary divergence, monitoring and analyzing the prices of more stocks would potentially lead to finding more trading opportunities. Here we report a stock pairs-trading system that finds trading opportunities for any two stocks in an $N$ -stock universe using a combinatorial optimization accelerator based on a quantum-inspired algorithm called simulated bifurcation. The trading opportunities are detected through solving an optimal path search problem in an $N$ -node directed graph with edge weights corresponding to the products of instantaneous price differences and statistical correlation factors between two stocks. The accelerator is one of Ising machines and operates consecutively to find multiple opportunities in a market situation with avoiding duplicate detections by a tabu search technique. It has been demonstrated in the Tokyo Stock Exchange that the FPGA (field-programmable gate array)-based trading system has a sufficiently low latency ( $33 \mu \text{s}$ for $N$ =15 or 210 pairs) to execute the pairs-trading strategy based on optimal path search in market graphs.

Financial portfolio construction problems are often formulated as quadratic and discrete (combinatorial) optimization that belong to the nondeterministic polynomial time (NP)-hard class in computational complexity theory. Ising machines are hardware devices that work in quantum-mechanical/quantum-inspired principles for quickly solving NP-hard optimization problems, which potentially enable making trading decisions based on NP-hard optimization in the time constraints for high-speed trading strategies. Here we report a real-time stock trading system that determines long(buying)/short(selling) positions through NP-hard portfolio optimization for improving the Sharpe ratio using an embedded Ising machine based on a quantum-inspired algorithm called simulated bifurcation. The Ising machine selects a balanced (delta-neutral) group of stocks from an $N$ -stock universe according to an objective function involving maximizing instantaneous expected returns defined as deviations from volume-weighted average prices and minimizing the summation of statistical correlation factors (for diversification). It has been demonstrated in the Tokyo Stock Exchange that the trading strategy based on NP-hard portfolio optimization for $N=128$ is executable with the FPGA (field-programmable gate array)-based trading system with a response latency of $164 \mu \text{s}$ .

In the design of ultralow leakage devices such as image sensors, it is necessary to understand the influence of low-density defects during plasma processing—plasma-induced physical damage (PPD)—on device performance. Defects created by plasma exposure act as carrier conduction sites and induce an increase in leakage (e.g., dark) current. This study proposes a PPD evaluation scheme for low-density defect assessments, specifically defects created in the lateral direction due to lateral stochastic straggling (lateral PPD). Two test structures were designed: a single device with a leakage current ( ${I} _{\mathrm{ leak}}$ ) and a complementary metal-oxide-semiconductor image sensor (CIS) circuit with a dark current ( ${I} _{\mathrm{ dark}}$ ). The energy level and density of defects distributed in the lateral direction were estimated using the Shockley–Read–Hall (SRH) model. The energy level ( ${E} _{\mathrm{ t}}$ ) was derived using the SRH model from the temperature dependence of ${I} _{\mathrm{ leak}}$ and ${I} _{\mathrm{ dark}}$ defined as the activation energy ${E} _{\mathrm{ a}}$ . The trap density ( ${N} _{\mathrm{ t}}$ ) was also determined. A comprehensive comparison of these parameters was conducted. Both ${I} _{\mathrm{ leak}}$ and ${I} _{\mathrm{ dark}}$ are dependent on the contact opening diameter after plasma exposure, which implies the presence of defects in the lateral direction via lateral PPD. From the analysis of the temperature dependence of ${I} _{\mathrm{ leak}}$ and ${I} _{\mathrm{ dark}}$ , the lateral PPD influenced the mean value of ${E} _{\mathrm{ a}}$ . Moreover, we confirmed that an increase in ${I} _{\mathrm{ dark}}$ indicates an increase in the number of trap sites, and more specifically, an increase in defects at shallow levels. The derived trap site density in the CIS circuit was consistent with that of a single device. The proposed evaluation scheme is useful for PPD evaluation in the presence of low-density defects, and is critical for the design of future low-leakage devices.

Low latency and reliable information transfer are highly demanded in fifth generation (5G) and beyond 5G wireless communications. A novel delay-bounded wireless media access control (MAC) protocol called Carrier Sense Multiple Access with Arbitration Point (CSMA/AP) was established to strictly ensure the upper boundary of communication delay. CSMA/AP enables collision-free and delay-bounded communications with a simple arbitration mechanism exploiting the precise time synchronization achieved by Wireless Two-Way Interferometry (Wi-Wi). Experimental demonstration and proving the feasibility in wireless environments are among the most critical steps before any further discussion of CSMA/AP and extension to various applications can take in place. In this work described in this paper, we experimentally demonstrated the fundamental principles of CSMA/AP by constructing a star-topology wireless network using software-defined radio terminals combined with precise time synchronization devices. We show that CSMA/AP was successfully operated, even with dynamic changes of the spatial position of the terminal or the capability to accommodate mobility, thanks to the real-time adaption to the dynamically changing environment by Wi-Wi. We also experimentally confirmed that the proposed CSMA/AP principle cannot be executed without Wi-Wi, which validates the importance of precise time synchronization. This study paves the way toward realizing delay-bounded wireless communications for future low-latency and highly reliable critical applications.

In this paper, we show that adversarial training time attacks by a few pixel modifications can cause undesirable overfitting in neural networks for both discriminative and generative models. We propose an evolutionary algorithm to search for an optimal pixel attack using a novel cost function inspired by domain adaptation literature to design our training time attack. The proposed cost function explicitly maximizes the generalization gap and domain divergence between clean and corrupted images. Empirical evaluations demonstrate that our adversarial training attack can achieve significantly low testing accuracy (with high training accuracy) on multiple datasets by just perturbing a single pixel in the training images. Even under the use of popular regularization techniques, we identify a significant performance drop compared to clean data training. Our attack is more successful than previous pixel-based training time attacks on state-of-the-art Convolutional Neural Networks (CNNs) architectures, as evidenced by significantly lower testing accuracy. Interestingly, we find that the choice of optimization plays an essential role in robustness against our attack. We empirically observe that Stochastic Gradient Descent (SGD) is resilient to the proposed adversarial training attack, different from adaptive optimization techniques such as the popular Adam optimizer. We identify that such vulnerabilities are caused due to over-reliance on the cross-entropy (CE) loss on highly predictive features. Therefore, we propose a robust loss function that maximizes the mutual information between latent features and input images, in addition to optimizing the CE loss. Finally, we show that the discriminator in Generative Adversarial Networks (GANs) can also be attacked by our proposed training time attack resulting in poor generative performance. Our paper is one of the first works to design attacks for generative models.

Planetary surface images are collected by sophisticated imaging devices onboard the orbiting spacecraft. Although these images enable scientists to discover and visualize the unknown, they often suffer from the `no-data' region because the data could not be acquired by the onboard instrument due to the limitation in operation time of the instrument and satellite orbiter control. This greatly reduces the usability of the captured data for scientific purposes. To alleviate this problem, in this article, we propose a machine learning-based `no-data' region prediction algorithm. Specifically, we leverage a deep convolutional neural network (CNN) based image inpainting algorithm to predict such unphotographed pixels in a context-aware fashion using adversarial learning on planetary images. The benefit of using our proposed method is to augment features in the unphotographed regions leading to better downstream tasks such as interesting landmark classification. We use the Moon and Mars orbital images captured by the JAXA's Kaguya mission and NASA's Mars Reconnaissance Orbiter (MRO) for experimental purposes and demonstrate that our method can fill in the unphotographed regions on the Moon and Mars images with good visual and perceptual quality as measured by improved PSNR and SSIM scores. Additionally, our image inpainting algorithm helps in improved feature learning for CNN-based landmark classification as evidenced by an improved F1-score of 0.88 compared to 0.83 on the original Mars dataset.COMP: Please replace colons appearing after figure numbers and table numbers with period in all figure and table captions.

The importance of reliable information transfer in wireless networks, especially regarding communication delay, is drastically increasing to fulfill safe and high-quality communication in the 5G and post-5G era. However, conventional media access control (MAC) protocol for wireless networks, notably Carrier Sense Multiple Access with Collision Avoidance (CSMA/CA), sometimes yields unexpectedly significant delay due to its complex arbitration mechanism assuming asynchronous communication among terminals. As the delay cannot be strictly bounded by a deterministic value, this causes a vulnerability of systems relying on wireless networks. This paper utilizes precise time synchronization achieved by Wireless Two-way Interferometry (Wi-Wi), enabling all terminals to be time-synchronized via wireless signals. We show that by an appropriate periodic assignment of each terminal’s data transmission timing, named Arbitration Point (AP), a simple arbitration algorithm obtains a strictly bounded maximum value for the delay while ensuring equalities among all participants. Furthermore, we demonstrate that the total number of terminals manageable in a star-topology wireless network significantly increases by densely packing AP timings, taking into account the spatial geometry information of terminals, which is another feature delivered by Wi-Wi measurement. In the meantime, we experimentally constructed a star-topology wired network where all terminals are time-synchronized via Wi-Wi to confirm the fundamental properties identified in the proposed arbitration protocol. This study paves a new way for future wireless networks where the delay is strictly bounded and provides the basis for ultra-reliable and high-quality information transfer functionalities by utilizing precise time synchronization and space localization (space-time synchronization).

Adaptive gradient methods such as adaptive moment estimation (Adam), RMSProp, and adaptive gradient (AdaGrad) use the temporal history of the gradient updates to improve the speed of convergence and reduce reliance on manual learning rate tuning, making them a popular choice for off-the-shelf Deep Neural Network (DNN) optimizers. In this article, we study the robustness of neural network optimizers in the presence of training perturbations. We show that popular adaptive optimization methods exhibit poor generalization while learning from noisy training data, compared to vanilla Stochastic Gradient Descent (SGD) and its variants, which manifest better implicit regularization properties. We construct an illustrative example of a family of two-class linearly separable toy-data such that models trained under noise using adaptive optimizers show only 52% test accuracy (random classifier), whereas SGD-based methods can achieve 100% test accuracy. We strengthen our hypothesis by empirical analysis using Convolutional Neural Networks (CNNs) on publicly available image datasets. For this purpose, our method trains neural network models with various optimizers on noisy training data, and we compute test accuracy on clean test data. Our results further highlight the robustness of SGD optimization against such noisy training data compared to its adaptive counterparts. Based on the results, our paper suggests a reconsideration of the extensive use of adaptive gradient methods for neural network optimization, especially when the training data is noisy.

We propose a multi-task learning framework for improving the performance of vision-based deep-learning approaches for driver distraction recognition. The most popular tool so far for solving this task is convolutional neural networks (CNNs) that have proven to be strongly biased toward local features. Such bias causes CNNs to neglect global structural information, adversely affecting the robustness of the distracted driver recognition task. To solve this problem, we generate positive and negative samples of each given input, and construct a triplet of images (i.e., raw image, positive sample, and negative sample). The positive sample is generated by applying structure-aware illumination to the human body region of each given input. The negative sample is generated by randomly shuffling the local regions of each given input. The networks are then trained with the triplets using a multi-task learning strategy to force the networks to explore global information by multiple tasks: (a) recognizing the raw input and positive sample as the given ground truth; (b) recognizing the negative sample as an extra “meaningless” label; (c) pulling closer the distance between the features obtained from the raw input and positive sample while pushing away the distance between the features obtained from the raw input and negative sample. By doing so, the model can be trained so that it neglects the background information and pays more attention to the global structual information of the scene. The proposed approach reaches state-of-the-art performance on the AUC Distracted Driver Dataset and performs better than state-of-the-art studies on the Drive and Act Dataset. With raw images as input, we have achieved an accuracy of 96.0% for the AUC distracted driver dataset and 66.8% for the Drive and Act Dataset. Our approach does not introduce extra overhead during the testing procedure (i.e., utilization procedure), which is helpful for real-life applications. Moreover, better accuracy can be achieved by fusing the predictions respectively obtained with the raw input and positive sample. As a result, we have achieved an accuracy of 96.3% for the AUC distracted driver Dataset and 66.9% for the Drive and Act Dataset. The class activation map (CAM) of our proposed method is subjectively more reasonable, which would enhance the reliability and explainability of the model.

Unilateral arm paralysis is a common symptom of stroke. In stroke patients, we observed that self-guided biomechanical support by the nonparetic arm unexpectedly triggered electromyographic activity with normal muscle synergies in the paretic arm. The muscle activities on the paretic arm became similar to the muscle activities on the nonparetic arm with self-supported exercises that were quantified by the similarity index (SI). Electromyogram (EMG) signals and functional near-infrared spectroscopy (fNIRS) of the patients (n=54) showed that self-supported exercise can have an immediate effect of improving the muscle activities by 40-80% according to SI quantification, and the muscle activities became much more similar to the muscle activities of the age-matched healthy subjects. Using this self-supported exercise, we investigated whether the recruitment of a patient's contralesional nervous system could reactivate their ipsilesional neural circuits and stimulate functional recovery. We proposed biofeedback training with self-supported exercise where the muscle activities were visualized to encourage the appropriate neural pathways for activating the muscles of the paretic arm. We developed the biofeedback system and tested the recovery speed with the patients (n=27) for 2 months. The clinical tests showed that self-support-based biofeedback training improved SI approximately by 40%, Stroke Impairment Assessment Set (SIAS) by 35%, and Functional Independence Measure (FIM) by 20%.

As part of their excretion care protocol, many of the senior citizens living in nursing homes are required to wear a diaper for urinary incontinence. However, the diaper must be changed by the caregiver after each incontinence episode, and some of the elderly affected may be unable to inform the caregiver about an episode because of cognitive problems or owing to a sense a shame. Being able to predict the volume of urine in the bladder without having to attach any sensors to the body would make it possible for the caregiver to take the residents to the toilet before they have an incontinence episode. Moreover, the residents would not have to wear a diaper. With this goal in mind, in this study, we developed a macroscopic model for urine accumulation in the bladder as well as a method for predicting the urine volume in the bladder based on the absorption spectrum of urine obtained immediately after urination. The proposed method does not involve attaching a sensor on the skin. We performed a series of experiments to evaluate the proposed method and calculated its error rate based on the actual amount of urine and that predicted using the method. The error rate was then compared with that for an ultrasonic sensor attached to the skin above the bladder. The mean of the error rate of the proposed method was 15.46% whereas that of the ultrasonic sensor was 23.42 %, with the two being significantly different at the 5% level.

We report low-loss optical Nyquist pulse train generation using a non-auxiliary wavelength selective switch (WSS). The typical approach for optical Nyquist pulse train generation involves two procedures. The first is conversion from a laser output Gaussian pulse to a Nyquist pulse via spectral filtering with a WSS. The second is multiplexing to generate a Nyquist pulse train with an optical circuit. To generate a high optical signal-to-noise ratio (OSNR) Nyquist pulse train, the first procedure was improved by developing a high-power laser and improving the filtering process by using nonlinear effects in a highly nonlinear fiber. The second procedure also has the potential to further improve the OSNR, because an optical circuit typically causes an optical loss of 9 dB in the case of eight-multiplexing. In this study, we demonstrate optical loss reduction for multiplexing using a nonauxiliary WSS approach without an auxiliary optical circuit. The experimental results show that the optical loss for the Nyquist eight-pulse train is successfully reduced to 2.8 dB.

High temperature characteristics of bipolar-mode operation of normally-off diamond junction field-effect transistors were investigated up to 573 K. As an important factor, the current gain depending on the gate current was analyzed with a theoretical model. We found that the experimental current gain decreased with the rise in the gate current, in agreement with the theoretical estimation considering the recombination at the end regions. We achieved 4-9 times higher drain currents in the bipolar-mode compared with the unipolar-mode operation at a DC current gain of 10. Furthermore, the bipolar-mode currents at the high temperatures of 473 and 573 K became two orders of magnitude larger than the unipolar-mode current at room temperature with a large DC current gain of 102.