Search

In this paper, we tackle the depth completion problem. Conventional depth sensors usually produce incomplete depth maps due to the property of surface reflection, especially for the window areas, metal surfaces, and object boundaries. However, we observe that the corresponding RGB images are still dense and preserve all of the useful structural information. The observation brings us to the question of whether we can borrow this structural information from RGB images to inpaint the corresponding incomplete depth maps. In this paper, we answer that question by proposing a Disentangling and Exchanging Network (DEN) for depth completion. The network is designed based on the assumption that after suitable feature disentanglement, RGB images and depth maps share a common domain for representing structural information. So we firstly disentangle both RGB and depth images into domain-invariant content parts, which contain structural information, and domain-specific style parts. Then, by exchanging the complete structural information extracted from the RGB image with incomplete information extracted from the depth map, we can generate the complete version of the depth map. Furthermore, to address the mixed-depth problem, a newly proposed depth representation is applied. By modeling depth estimation as a classification problem coupled with coefficient estimation, blurry edges are enhanced in the depth map. At last, we have implemented ablation experiments to verify the effectiveness of the proposed DEN model. The results also demonstrate the superiority of DEN over some state-of-the-art approaches.

Nonvolatile computing-in-memory (nvCIM) can improve the latency (t AC ) and energy-efficiency (EF MAC ) of tiny AI edge devices performing multiply-and-accumulate (MAC) computing after system wake-up. Prior nvCIMs have proven effective for binary input (IN) and weight (W), and 3b output (OUT) [1], 1-8-1b IN-W-OUT [2], and 2-3-4b IN-W-OUT [3] neural networks; however, the higher precision (4-4b IN-W) for MAC operations is needed for multi-bit CNNs to achieved high-inference accuracy [4]. As Fig.15.4.1 shows, improving the precision of nvCIM macros involves various challenges. (1) A large number of activated WLs provides a wide range of BL current (I BL ) resulting in an inaccurate BL-clamping voltage (V BLC ); as well as a large (I BL ) requiring a large array area due to the need for wide metal lines to support high-current density. (2) Previous “WL = input” approaches suffer from: (a) few parallel inputs (IN#) due to (1), and (b) long (t AC ) in multiple cycles of binary WL inputs on 1T1R cells for multibit inputs. (3) Previous positive-negative-split weight-mapping consumes high total (l BL ) and area overhead (needing 2x(m-1) cells for a signed m-bit weight) for cell arrays with high-weight precision. (4) Long (t AC ) and a large number of reference currents (IREF#) for high-precision outputs. To overcome these challenges, this work proposes: (1) a BL-IN-OUT multibit computing (BLIOMC) scheme using a single WL-on and input-aware multibit BL clamping (IA-MBC) to shorten (l BL ) for multibit inputs, increase IN#, and reduce the (l BL ) range/size for accurate (V BLC ) and a compact array area. (2) Scrambled 2's complement (S2C) weight mapping (S2CWM), input-aware source-line (SL) voltage biasing (IA-SLVB), and an S2C value combiner (S2CVC) to reduce area overhead and l BL in the cell array. (3) A dual-bit small-offset current-mode sense amplifier (DbSO-CSA) to reduce IREF# and t AC . A fabricated 22nm 2Mb ReRAM-CIM macro presents the first 4b-input nvCIM macro, featuring a 9.8-18.3ns t AC and an EF MAC of 121.3-28.9TOPS/W from binary to 4bIN-4bW-11bOUT compute precisions.

Many Al edge devices require local intelligence to achieve fast computing time (t AC ), high energy efficiency (EF), and privacy. The transfer-learning approach is a popular solution for Al edge chips, wherein data used to re-train the Al in the cloud is used to fine-tune (re-train) a few of the neural layers in edge devices. This enables the dynamic incorporation of data from in-situ environments or private information. Computing-in-memory (CIM) is a promising approach to improve EF for Al edge chips, existing CIM schemes support inference [1]–[5] with forward (FWD) propagation; however, they do not support training, requiring both FWD and backward (BWD) propagation, due to differences in weight-access flow for FWD and BWD propagation. As Fig. 15.2.1 shows, efforts to increase the precision of the input (IN), weight (W), and/or output (OUT) tend to degrade r AC and EF for training operations irrespective of scheme: digital FWD and BWD (DF-DB) or CIM-FWD-digital-BWD (CiMF-DB). This work develops a two-way transpose (TWT) SRAM-CIM macro supporting multibit MAC operations for FWD and BWD propagation with fast r AC and high EF within a compact area. The proposed scheme features (1) A TWT multiply cell (TWT-MC) with a high resistance to process variation; and (2) a small-offset gain-enhancement sense amplifier (SOGE-SA) to tolerate a small read margin. A 28nm 64Kb TWT SRAM-CIM macro was fabricated using a foundry-provided compact 6T-SRAM cell for SRAM-CIM devices supporting both inference and training operations for the first time. This macro also demonstrates the fastest t AC (3.8 – 21ns) and highest EF (7 – 61.1TOPS/w) for MAC operations using 2 – 8b inputs, 4 – 8b weights and 12 − 20b outputs.

Developing cost-effective and fast detection methods in full-field optical coherence microscopy (FF-OCM) platform for biological applications is of prime focus today. In this paper, we report a cost-effective Linnik type FF-OCM system with a single broadband white light emitting diode (WLED) (470-850 nm, λ0 = 630 nm) light source in conjunction with a derivative-based image reconstruction algorithm. Sequential 2-D multiple phase-shifted interferograms were obtained by moving the sample stage using piezoelectric transducer and recorded via a 2-D charge complementary oxide semiconductor camera. Subsequently, the en-face sectional images were reconstructed using a derivative-based algorithm. The measured axial and lateral resolutions along with interferometric phase stability of the system were 0.9 μm, 1.4 μm, and 11 mrad, respectively. The advantages of the derivative-based algorithm over 4-step phase shifted method were demonstrated using USAF target and onion as test samples. Moreover, studies via the present system of the microstructure of skin epidermis in mouse ear with melanoma showed a significant deterioration from normal mouse skin in stratum corneum of the epidermis. This is the first study of the application of a derivative-based algorithm for biological samples in FF-OCM wherein WLED is used as a light source. The present system is stable, compact, and cost-effective compared to the conventional FF-OCM systems and provides comparable spatial resolution.

The restriction of access to software systems is more important than ever. For example, critical data is increasingly being stored on web services that are accessible from anywhere in the world. Yet most primary authentication methods are still largely based on passwords, which are vulnerable to various attacks such as phishing scams and keyloggers. Advanced methods of behavior-based authentication exist, but most are designed for a specific area or system and are not generally applicable. In this paper, we propose a generic continuous authentication scheme for software systems, which supplements existing authentication schemes and works as an auxiliary layer to provide additional protection against impostors. The kernel of our scheme is a novel monitoring engine that detects impostors in real-time based on behavior and context information. We evaluate our scheme on a dataset consisting of real users' historical records provided by our industrial partner, and the results demonstrate that our approach achieves a high classification accuracy with only a short delay in detection, allowing for real-time, continuous authentication.

Randomness is a significant ingredient to achieve fairness on many applications (i.e., gambling). On the use of blockchain and smart contract, we observe that some existing lottery and gambling games involve one single source to produce randomness. As known, Fomo3D uses the game state to determine the winner, whereas TrueFlip uses the blockchain state (at a specific point in time). Those system do not rely on significant randomness extraction. In this paper, we present a lottery DApp and implement it by using smart contracts. In this DApp, the randomness is determined not only from game and blockchain states, but also from committees.

Recently, many researches are focusing on millimeter wave (mmWave) due to its large bandwidth resources. In the next 5G generation, wireless multi- gigabit (WiGig) is developed based on IEEE 802.11ad/ay standards at unlicensed mmWave bands for providing extremely high throughput. However, mmWave has high propagation loss due to high frequency transmission properties. Beamforming (BF) technique is adopted to solve this problem, and therefore, we have to perform beam training before data transmission. In the standard, the conventional method for beam training is exhaustive beam search (EBS) which takes too much time so that the data transmission time will decrease. On the other hand, blocked environments may severely degrade WiGig beam training performance, and most of existing algorithms do not consider this issue. Recently, machine and deep learning have been widely used in the wireless communication field. We propose a learning-based beam training (LBT) to simultaneously learn about wireless environments and beam training candidates. We select simplified neural network (NN) model to achieve lower computation overhead. To further refine learning information, we propose two enhanced algorithm, expanded LBT (LBT-E) and history-aided expanded LBT(LBT-HE). LBT-E aims to tackle unexpected slight deviation and LBT- HE make use of historical information to improve beam matching accuracy with acceptable latency. In our simulation, our proposed learningbased schemes achieve much higher throughput compared to EBS and algorithms in existing literatures.

Physically unclonable functions (PUFs) are promising primitives for hardware security with wide applications in the lnternet of Things (IoT), e.g., authentication and encryption key generation [1, 2]. Most silicon PUFs utilize process variability of semiconductor manufacturing [1, 3, 4]. These implementations are sensitive to variations in operating conditions (e.g., supply voltage and temperature variations) and undergo significant native bit-error-rates (N-BERs). Thus, additional stabilizing strategies, such as ECC, majority voting, and masking, are necessary. Furthermore, the PUF key after enrollment cannot be changed in prior implementations [1–5]. This could be unsafe if the PUFs are repeatedly used in insecure environments, as PUFs suffer from the challenges of ownership change and overuse (Fig. 25.2.1).

Embedded nonvolatile memory (NVM) and computing-in-memory (CIM) are significantly reducing the latency (t MAC ) and energy consumption (E MAC ) of multiply- and-accumulate (MAC) operations in artificial intelligence (AI) edge devices [1, 2]. Previous ReRAM CIM macros demonstrated MAC operations for lb-input, ternary- weighted, 3b-output CNNs [1] or lb-input, 8b-weighted, 1b-output fully-connected networks with limited accuracy [2]. To support higher-accuracy convolution neural network heavy applications NVM-CIM should support multibit inputs/weights and multi-bit output (MAC-OUT) for CNN operations. One way to achieve multibit weights is to use a multi-level ReRAM cell to store the weight. However, as shown in Fig. 24.1.1, multibit ReRAM CIM faces several challenges. (1) a tradeoff between area and speed for multibit input/weight/MAC-OUT MAC operations; (2) sense amplifier’s high input offset, large area, and high parasitic load on the read-path due to large BL currents (I BL ) from multibit MAC; (3) limited accuracy due to a small read/sensing margin (I SM ) across MAC-OUT or variation in cell resistance (particularly MLC cells). To overcome these challenges, this work proposes, (1) a serial-input non-weighted product (SINWP) structure to optimize the tradeoff between area, t MAC and E MAC , (2) a down-scaling weighted current translator (DSWCT) and positive-negative current- subtractor (PN-ISUB) for short delay, a small offset and a compact read-path area; and (3) a triple-margin small-offset current-mode sense amplifier (TMCSA) to tolerate a small I SM . A fabricated 55nm 1Mb ReRAM-CIM macro is the first ReRAM CIM macro to support CNN operations using multibit input/weight MAC-OUT. This device achieves the shortest CIM-MAC-access time (t AC ) among existing ReRAM-CIMs (t MAC =14.6ns with 2b-input, 3b-weight with 4b-MAC-OUT) and the best peak E MAC of 53.17 TOPS/W (in binary mode).

Demodex mite is believed to cause skin diseases such as rosacea, demodex folliculitis and demodex-aggravated perioral dermatitis. Its conventional diagnostic methods are skin scrape tests and superficial biopsies, which are invasive and painful to patients.

A full-field optical coherence tomography with a swept source integrating a superluminescent diode and an acousto-optic tunable filter is designed, implemented, and tested. The center wavelength and the bandwidth of the swept source are 840nm and 50nm, respectively. There are estimated about 688 laser lines generated in a sweep cycle. For convenience, a Mirau objective lens is adopted to simplify the system structure. Some preliminary experiments were conducted by using the system, and an onion was adopted as the sample. Two clear restored 3D images of the onion were obtained by using two different algorithms. The differences between the two images were also compared with each other. Further improvement and experiments by using this system are still in progress.

In this investigation Zr-based (Zr 51 Cu 31 Al 3 Ni 5 ) and Ti-based (Ti 40 Zr 10 Cu 8 Nb 7 Co 7 ) thin-film metallic glass (TFMG), in different thickness, were used to improve the tribological properties of 7075 aluminum alloy substrates with different surface patter. Wear tests were performed in refrigerator oil to evaluate the effect of thin film adhesion strength and surface pattern on the wear resistance. A scanning electron microscope (SEM) with energy dispersive spectrometer (EDS), nanoindenter and scratch tester were used to analyze the metallurgical and mechanical properties of the specimens. The stress distribution of TFMG on the 7075 aluminum alloy substrates under line contact condition was evaluated by ANSYS. According to the experiment results, wear resistance of the sample that was coated with a 200nm-thick Ti-based was better than the other coated samples, because surface texture can increase mechanical locking effect to enhance adhesive strength, and Ti-based thin film coating can improve tribological performance and reduce wear loss.

In this paper we describe a scheduling framework that allocates resources to both batch jobs and interactive jobs simultaneously in a private cloud with a static amount of resources. In the system, every job has an individual service level agreement (SLA), and violating the SLA incurs penalty. We propose a model to formally quantify the SLA violation penalty of both batch and interactive jobs. The analysis on the interactive jobs focuses on queuing analysis and response time. The analysis on batch jobs focuses on the non-preemptive job scheduling for multiple processing units. Based on this model we also propose algorithms to estimate the penalty for both batch jobs and interactive jobs, and algorithms that reduce the total SLA violation penalty. Our experiment results suggest that our system effectively reduces the total penalty by allocating the right amount of resources to heterogeneous jobs in a private cloud system.

Utilizing advanced technologies to guarantee safety protection for pedestrians at intersections has become an important issue because intersections are the places where pedestrians easily conflict with vehicles. The purpose of this study is to propose a conceptual design of an intersection bus-pedestrian collision warning system with appropriate detection and warning parameters for bus drivers approaching an intersection. Based on the design concept, bus drivers' perception-reaction time, emergency deceleration rate of buses, and pedestrian walking speed are defined as the basic parameters. A bus driving simulation is designed and conducted to collect bus drivers' responses to the suddenly crossing pedestrians at unsignalized intersections or signalized intersections with green interval for parameters analysis. Finally, the warning timings for auditory warnings and visual warnings, the locations for vehicle detectors and pedestrian detectors, and the locations for visual warning devices are developed through a further analysis of the experimental results.

Radio Frequency Identification (RFID) is kind of automatic wireless recognition sum according to the gain technology, its special capability of data chasing and data collecting can load messages directly to the database and deal with many labels at one time. RFID has been applied in many domains, including medical application, like medical record system. The major of this study is medical application, this study expects RFID can combine with the data of health examination, and sort the health conditions, so that doctors and medical advisers can provide the medical information or ways of health case immediately.The purpose of this study is to classify the data of health examination we got by data mining and RFID’s special capability of chasing and collecting. This study expects that we can classify the health condition of participants of health examination by the data of health examination into three classes (high risk, median risk, low risk), and provide different medical service to people by different classes.

In this paper, we proposed a low complexity subdivision algorithm to approximate Phong shading. It is a combination of a subdivision scheme using forward difference and a recovery scheme to prevent rasterization anomaly. Dual space subdivision, triangle filtering and variable sharing schemes are also proposed to reduce the computation. Compared with the conventional recursive subdivision shading algorithm, the proposed algorithm reduces 46.6% memory access, 80% projective transformations and perspective division, 93.75% clipping/culling tests and 60% 3x3 matrix multiplications for setup operation.

Wireless Sensor Networks (WSNs) is composed of sink and many sensor nodes. It has the great scope to monitor environment and the characteristics of wireless communications. Relatively, the advantages of sensor node bring themselves some restrictions. The greatest challenge among those is the constraints on energy. The most critical issue is how to minimize the energy consumption for keeping target tracking to devise the mobile target tracking mechanism in WSNs.Therefore, we propose a solution on the basis of the advantages of Grid-based network which could determine where the target might be in the grids, and then those sensor node do track the target; others turn into sleep mode, this mechanism could efficiently reduce the energy consumption, and even prolongs the whole network’s lifetime.

To build an automatic assistance for learning and provide a self-paced learning mechanism are the objectives in the e-learning environment. The paper improves the previous proposed Semantic Question Answering System which applies Link Grammar and WordNet to form a Semantic Tree to represent the meaning of question and further find the relevant answer based on the selected Data Structure Ontology. The paper addresses the improved functions including the flexible answering method which can refine the Semantic Tree to find more relevant answers, the Ontology Extension Module is designed to acquire data from Internet to raise the ontology and the Feedback for System module is designed for the instructors to provide more sufficient knowledge.

Since the mass development of the wireless network, the advancement in wireless communications and micro-electronics element promote the development of wireless sensor networks (WSNs). The sensor nodes of WSNs have the following abilities such as: sensing, computation, and wireless communication capabilities. Hence, WSNs can be used for various application domains. Recently, the localization issue in the WSNs has become an important research topic. In this paper, we propose a location mechanism with mobile reference nodes and Global Positioning System (GPS) to help sensor nodes to calculate the coordinate location information. And then sensor nodes deliver the coordinate location to base station through neighboring sensor nodes. By using this mechanism, we can calculate the coordinate location of the target sensor nodes. And then we can apply this mechanism to track or transfer data packets in the WSNs efficiently.