Search

We tackle the problem of class incremental learning (CIL) in the realm of landcover classification from optical remote sensing (RS) images in this paper. The paradigm of CIL has recently gained much prominence given the fact that data are generally obtained in a sequential manner for real-world phenomenon. However, CIL has not been extensively considered yet in the domain of RS irrespective of the fact that the satellites tend to discover new classes at different geographical locations temporally. With this motivation, we propose a novel CIL framework inspired by the recent success of replay-memory based approaches and tackling two of their shortcomings. In order to reduce the effect of catastrophic forgetting of the old classes when a new stream arrives, we learn a curriculum of the new classes based on their similarity with the old classes. This is found to limit the degree of forgetting substantially. Next while constructing the replay memory, instead of randomly selecting samples from the old streams, we propose a sample selection strategy which ensures the selection of highly confident samples so as to reduce the effects of noise. We observe a sharp improvement in the CIL performance with the proposed components. Experimental results on the benchmark NWPU-RESISC45, PatternNet, and EuroSAT datasets confirm that our method offers improved stability-plasticity trade-off than the literature.

In his paper, we present order reduction techniques for nonlinear quasi-periodic systems subjected to external excitations. The order reduction techniques presented here are based on the Lyapunov-Perrone (L-P) Transformation. For a class of non-resonant quasi-periodic systems, the L-P transformation can convert a linear quasi-periodic system into a linear time-invariant one. This Linear Time-Invariant (LTI) system retains the dynamics of the original quasi-periodic system. Once this LTI system is obtained, the tools and techniques available for analysis of LTI systems can be used, and the results could be obtained for the original quasi-periodic system via the L-P transformation. This approach is similar to using the Lyapunov-Floquet (L-F) transformation to convert a linear time-periodic system into an LTI system and perform analysis and control. Order reduction is a systematic way of constructing dynamical system models with relatively smaller states that accurately retain large-scale systems' essential dynamics. In this work, reduced-order modeling techniques for nonlinear quasi-periodic systems subjected to external excitations are presented. The methods proposed here use the L-P transformation that makes the linear part of transformed equations time-invariant. In this work, two order reduction techniques are suggested. The first method is simply an application of the well-known Guyan like reduction method to nonlinear systems. The second technique is based on the concept of an invariant manifold for quasi-periodic systems. The 'quasi-periodic invariant manifold' based technique yields' reducibility conditions.' These conditions help us to understand the various types of resonant interactions in the system. These resonances indicate energy interactions between the system states, nonlinearity, and external excitation.

In this paper, we propose a novel framework for multi-image co-segmentation using class agnostic meta-learning strategy by generalizing to new classes given only a small number of training samples for each new class. We have developed a novel encoder-decoder network termed as DVICE (Directed Variational Inference Cross Encoder), which learns a continuous embedding space to ensure better similarity learning. We employ a combination of the proposed DVICE network and a novel few-shot learning approach to tackle the small sample size problem encountered in co-segmentation with small datasets like iCoseg and MSRC. Furthermore, the proposed framework does not use any semantic class labels and is entirely class agnostic. Through exhaustive experimentation over multiple datasets using only a small volume of training data, we have demonstrated that our approach outperforms all existing state-of-the-art techniques., Comment: Accepted at 2020 25th International Conference on Pattern Recognition (ICPR)

Monolithic Active Pixel Sensors (MAPS) have been developed since the late 1990s employing silicon substrate with a thin epitaxial layer in which deposited charge is collected by disordered diffusion rather than by drift in an electric field. As a consequence the signal is small and slow, and the radiation tolerance is below the requirements for LHC experiments by factors of 100 to 1000. We developed fully depleted (D)MAPS pixel sensors employing a 150 nm CMOS technology and using a high resistivity substrate as well as a high biasing voltage. The development has been carried out in three subsequent iterations, from prototypes to a large pixel matrix comprising a complete readout architecture suitable for LHC operation. Full CMOS electronics is embedded in large deep n-wells which at the same time serve as collection nodes (large electrode design). The devices have been intensively characterized before and after irradiation employing lab tests as well as particle beams. The devices can cope with particle rates seen by the innermost pixel detectors of the LHC pp-experiments or as seen by the outer pixel layers of the planned HL-LHC upgrade. They are radiation hard to particle fluences of at least $10^{15}~\mathrm{n_{eq}/cm^2}$ and total ionization doses of at least 50 Mrad.

Depleted Monolithic Active Pixel Sensor (DMAPS) prototypes developed in the TowerJazz 180 nm CMOS imaging process have been designed in the context of the ATLAS upgrade Phase-II at the HL-LHC. The pixel sensors are characterized by a small collection electrode (3 $\mu$m) to minimize capacitance, a small pixel size ($36.4\times 36.4$ $\mu$m), and are produced on high resistivity epitaxial p-type silicon. The design targets a radiation hardness of $1\times10^{15}$ 1 MeV n$_{eq}$/cm$^{2}$, compatible with the outermost layer of the ATLAS ITK Pixel detector. This paper presents the results from characterization in particle beam tests of the Mini-MALTA prototype that implements a mask change or an additional implant to address the inefficiencies on the pixel edges. Results show full efficiency after a dose of $1\times10^{15}$ 1 MeV n$_{eq}$/cm$^{2}$.

We propose that Cu-substituted Fe$_2$P, (Fe$_{1-x}$Cu$_x$)$_2$P ($x\sim 0.16$), to be an outstanding contender for the STT-MRAM application. Using first principles based calculations in the framework of density functional theory and through Monte Carlo simulations, we demonstrate that this material can be used as ferromagnetic electrode in the magnetic tunnel junction (MTJ) of STT-MRAM due to its moderate perpendicular magnetic anisotropy (PMA), large tunnel magneto-resistance (TMR), good thermal stability and high ferromagnetic transition temperature. We point out that the simplicity in the synthesis, huge abundance, and non-toxicity make this material a very good candidate to replace the current MTJ materials for STT-MRAM such as FePt,~FeCo or FeCoB.

We find that the temperature dependence of the Electron Paramagnetic Resonance (EPR) linewidth observed in Bi$_{0.5}$Sr$_{0.5}$Mn$_{0.9}$Cr$_{0.1}$O$_3$ (BSMCO), as well as in certain other 3-dimensional manganites undergoing antiferromagnetic transitions is satisfactorily described by the Berezinskii-Kosterlitz-Thouless (BKT) model. We understand this unexpected result in terms of an effective 2-dimensional XY easy plane anisotropy induced by the magnetic field applied in the EPR experiment. This conclusion is supported by the field dependence of the BKT correlations observed in the quasi two-dimensional antiferromagnetic compound BaNi$_{2}$V$_{2}$O$_{8}$.

We study the effects of 10% Cr substitution in Mn sites of Bi0.5Sr0.5MnO3 on the antiferromagnetic (AFM) (TN ~ 110 K) transition using structural, magnetic and electron paramagnetic resonance (EPR) techniques. Field cooled (FC) and zero field cooled (ZFC) magnetization measurements done from 400 K down to 4 K show that the compound is in the paramagnetic (PM) phase till 50 K where it undergoes a transition to a short range ferromagnetic phase (FM). Electron paramagnetic resonance measurements performed in the temperature range 300 K till 80 K conform with the magnetization measurements as symmetric signals are observed owing to the paramagnetic phase. Below 80 K, signals become asymmetric. Electron paramagnetic resonance intensity peaks at ~ 110 K, the decreasing intensity below this temperature confirming the presence of antiferromagnetism. We conclude that below 50 K the magnetization and EPR results are consistent with a cluster glass phase of BSMCO, where ferromagnetic clusters coexist with an antiferromagnetic background.

This work presents a depleted monolithic active pixel sensor (DMAPS) prototype manufactured in the LFoundry 150\,nm CMOS process. DMAPS exploit high voltage and/or high resistivity inclusion of modern CMOS technologies to achieve substantial depletion in the sensing volume. The described device, named LF-Monopix, was designed as a proof of concept of a fully monolithic sensor capable of operating in the environment of outer layers of the ATLAS Inner Tracker upgrade in 2025 for the High Luminosity Large Hadron Collider (HL-LHC). This type of devices has a lower production cost and lower material budget compared to presently used hybrid designs. In this work, the chip architecture will be described followed by the characterization of the different pre-amplifier and discriminator flavors with an external injection signal and an iron source (5.9\,keV x-rays)., Comment: 2 pages, 3 figures, proceedings of the 14th Pisa Meeting on Advanced Detectors (PM2018), submitted to Nucl. Instr. Meth. A