Search

Continual learning (CL) refers to the ability of an algorithm to continuously and incrementally acquire new knowledge from its environment while retaining previously learned information. A model trained on one data modality often fails when tested with a different modality. A straightforward approach might be to fuse the two modalities by concatenating their features and training the model on the fused data. However, this requires retraining the model from scratch each time it encounters a new domain. In this paper, we introduce a continual learning algorithm capable of incrementally learning different data modalities by leveraging both class-incremental and domain-incremental learning scenarios in an artificial environment where labeled data is scarce, yet non-iid (independent and identical distribution) unlabeled data from the environment is plentiful. The proposed algorithm is efficient and only requires storing prototypes for each class. We evaluate the algorithm's effectiveness on a challenging custom multimodal dataset comprising of tactile data from a soft pneumatic gripper, and visual data from non-stationary images of objects extracted from video sequences. Additionally, we conduct an ablation study on the custom dataset and the Core50 dataset to highlight the contributions of different components of the algorithm. To further demonstrate the robustness of the algorithm, we perform a real-time experiment for object classification using the soft gripper and an external independent camera setup, all synchronized with the Robot Operating System (ROS) framework., Comment: 15 pages, 10 figures

Machine learning (ML) is no new concept in the high-energy physics community, in fact, many ML techniques have been employed since the early 80s to deal with a broad spectrum of physics problems. In this paper, we present a novel technique to separate electrons from pions in the Gas Electron Multiplier Transition Radiation Detector (GEM TRD) using deep learning. The Artificial Neural Network (ANN) model is trained on the Monte Carlo data simulated using the ATHENA-based detector and simulation framework for the Electron-Ion Collider (EIC) experiment. The ANN model does a good job of separating electrons from pions.

ATHENA has been designed as a general purpose detector capable of delivering the full scientific scope of the Electron-Ion Collider. Careful technology choices provide fine tracking and momentum resolution, high performance electromagnetic and hadronic calorimetry, hadron identification over a wide kinematic range, and near-complete hermeticity. This article describes the detector design and its expected performance in the most relevant physics channels. It includes an evaluation of detector technology choices, the technical challenges to realizing the detector and the R&D required to meet those challenges.

Spiking Neural Networks (SNN) are energy-efficient computing architectures that exchange spikes for processing information, unlike classical Artificial Neural Networks (ANN). Due to this, SNNs are better suited for real-life deployments. However, similar to ANNs, SNNs also benefit from deeper architectures to obtain improved performance. Furthermore, like the deep ANNs, the memory, compute and power requirements of SNNs also increase with model size, and model compression becomes a necessity. Knowledge distillation is a model compression technique that enables transferring the learning of a large machine learning model to a smaller model with minimal loss in performance. In this paper, we propose techniques for knowledge distillation in spiking neural networks for the task of image classification. We present ways to distill spikes from a larger SNN, also called the teacher network, to a smaller one, also called the student network, while minimally impacting the classification accuracy. We demonstrate the effectiveness of the proposed method with detailed experiments on three standard datasets while proposing novel distillation methodologies and loss functions. We also present a multi-stage knowledge distillation technique for SNNs using an intermediate network to obtain higher performance from the student network. Our approach is expected to open up new avenues for deploying high performing large SNN models on resource-constrained hardware platforms., Comment: Preprint: Manuscript under review

In this paper, we have used the QUICK scheme of the finite volume method to investigate the problem of steady 2-D free convective incompressible flow with heat and mass transfer in an inclined rectangular domain at different Rayleigh numbers in the range of $10^4 \le Ra \le 10^6$, for Prandtl number $Pr=7.2$, and Lewis number $Le=1$. We have used no-slip wall boundary conditions for the components of velocity and Neumann boundary conditions for temperature and concentration. We have used the QUICK scheme to discretize the governing equations along with the boundary conditions chosen in the present problem. The SIMPLE algorithm is adopted to compute the numerical solutions of flow variables, $U$-velocity, $V$-velocity, pressure, temperature, and concentration as well as the local and average Nusselt and Sherwood numbers at different Rayleigh numbers in the range mentioned above. Our numerical solutions for $U$-velocity along the vertical line through the geometric center has been compared with benchmark solutions available in the literature for $\lambda=15^{\circ},\, Ra=10^4,\, Pr=7.2$, and $Le=1$ and it has been found that it is in the good agreement. Pressure increases with increasing the angles of inclination. When the angles of inclination increase from $\lambda=15^{\circ}$ to $\lambda=45^{\circ}$, the intensity of streamlines decreases near the center and secondary cells are formed at the top and bottom of the center of the domain. As the Rayleigh number is increased from $10^4$ to $10^6$, the average Nusselt number is decreased for $\lambda=15^{\circ}$, and decreases and then increases for $\lambda=45^{\circ}$, whereas, the average Sherwood number increases for $\lambda=15^{\circ}$ and decreases for $\lambda=45^{\circ}$.

In this paper, we have proposed a modified Marker-And-Cell (MAC) method to investigate the problem of an unsteady 2-D incompressible flow with heat and mass transfer at low, moderate, and high Reynolds numbers with no-slip and slip boundary conditions. We have used this method to solve the governing equations along with the boundary conditions and thereby to compute the flow variables, viz. $u$-velocity, $v$-velocity, $P$, $T$, and $C$. We have used the staggered grid approach of this method to discretize the governing equations of the problem. A modified MAC algorithm was proposed and used to compute the numerical solutions of the flow variables for Reynolds numbers $Re = 10$, 500, and 50,000 in consonance with low, moderate, and high Reynolds numbers. We have also used appropriate Prandtl $(Pr)$ and Schmidt $(Sc)$ numbers in consistence with relevancy of the physical problem considered. We have executed this modified MAC algorithm with the aid of a computer program developed and run in C compiler. We have also computed numerical solutions of local Nusselt $(Nu)$ and Sherwood $(Sh)$ numbers along the horizontal line through the geometric center at low, moderate, and high Reynolds numbers for fixed $Pr = 6.62$ and $Sc = 340$ for two grid systems at time $t = 0.0001s$. Our numerical solutions for u and v velocities along the vertical and horizontal line through the geometric center of the square cavity for $Re = 100$ has been compared with benchmark solutions available in the literature and it has been found that they are in good agreement. The present numerical results indicate that, as we move along the horizontal line through the geometric center of the domain, we observed that, the heat and mass transfer decreases up to the geometric center. It, then, increases symmetrically.

Numerical simulations for 2-D unsteady, incompressible flow with heat transfer in a four-sided lid-driven rectangular domain are reported in the present study. For the four-sided lid-driven rectangular domain, the lower wall is moved to the left, the upper wall is moved to the right, while the right wall is moved upwards and the left wall is moved downwards. All four walls move with equal speed. Different constant temperatures are applied to the left and right moving walls, and thermal insulation is applied to the upper and bottom moving walls. The governing equations are discretized using the QUICK scheme of finite volume methods. The SIMPLE algorithm is adopted to compute the numerical solutions of the flow variables, $u$-velocity, $v$-velocity, $P$, and $\theta$ as well as local and average Nusselt numbers for $50 \le Re \le 1500$ and $Pr=6.63$. Due to the force generated by moving fluid, the direction of moving walls and the Reynolds number affect fluid flow in the rectangular domain in addition, at different Reynolds numbers along the cold wall of the domain, the variation in average and local Nusselt numbers reveals that overall heat transfer increases isotherms showed that as Reynolds numbers increase, the horizontal temperature gradient near the vertical walls decreases, because of which heat transfer decreases.decreases., Comment: 10 pages, 8 figures