박정우 교수

Advances in neural network and machine learning algorithms have sparked a wide array of research in specialized hardware, ranging from high-performance convolutional neural network (CNN) accelerators to energy-efficient deep-neural network (DNN) edge computing systems [1]. While most studies have focused on designing inference engines, recent works have shown that on-chip training could serve practical purposes such as compensating for process variations of in-memory computing [2] or adapting to

Recent advances in neural network (NN) and machine learning algorithms have sparked a wide array of research in specialized hardware, ranging from high-performance NN accelerators for use inside the server systems to energy-efficient edge computing systems. While most of these studies have focused on designing inference engines, implementing the training process of an NN for energy-constrained mobile devices has remained to be a challenge due to the requirement of higher numerical precision. In

Recent works on mobile deep-learning processors have presented designs that exploit sparsity [2, 3], which is commonly found in various neural networks. However, due to the shift in the machine learning community towards using non-sparse activation functions such as Leaky ReLU or Swish for better training convergence, state-of-theart models no longer exhibit the sparsity found in conventional ReLU-based models (Fig. 9.3.1, top). Moreover, contrary to error-tolerant image classification tasks, mo

Modern SoCs have the large scale and complexity. Modeling hardware architectures for SoCs usually requires pin-level and cycle-level descriptions by register transfer level(RTL). Hardware design at RTL takes too much effort to develop and simulate HDL code such as Verilog or VHDL. Therefore, a higher abstraction level is needed. Transaction level modeling (TLM) using SystemC addresses the limitations of pure RTL modeling methodologies. In this paper, we apply co-simulation of SystemC TLM with RT

This paper presents an electrical impedance tomography (EIT) method using a partial-differential-equation-constrained optimization approach. The forward problem in the inversion framework is described by a complete electrode model (CEM), which seeks the electric potential within the domain and at surface electrodes considering the contact impedance between them. The finite element solution of the electric potential has been validated using a commercial code. The inverse medium problem for recons

This paper proposes a low-complexity mobile display digital interface (MDDI) architecture. We implement and verify the proposed MDDI-host architecture on a system-on-chip (SoC) platform with a 32-bit RISC CPU and a FPGA module. As a result of verification, the gate count of the MDDI host is reduced by about 50 % compared to that of the conventional parallel interface. And the bandwidth per pin of the MDDI host is 3 times more than that of a conventional parallel interface. Furthermore, the MDDI

This study discusses a nonlinear electrical impedance tomography (EIT) technique under different analysis conditions to propose its optimal implementation parameters. The forward problem for calculating electric potential is defined by the complete electrode model. The inverse problem for reconstructing the target electrical conductivity profile is presented based on a partial-differential-equation-constrained optimization approach. The electrical conductivity profile is iteratively updated by s

과학적 모형은 특정한 물리적 현상을 기술, 설명, 예측할 수 있는 개념 체계를 말한다. 과학적 모형의 사회적 구성 수업은 과학교육 분야에서 새로운 교수 학습 전략으로 주목받고 있으며 다양한 연구가 진행되고 있다. 모형을 통한 예상과 실제 세계에서 얻은 자료와의 일치, 불일치에 따라 모형의 적합성을 판단하고 모형...

While the recent development of high performance mixedreality (MR) devices is enabling its use in medical and industrial domains, this requires hand gesture recognition to be robust to different textures inflicted by gloves often worn for hygiene and safety purposes. Unfortunately, most existing hand gesture datasets are not captured using recent commercial MR devices, and none addresses the issue of wearing gloves in gesture recognition. We aim to fill these gaps by introducing a new dataset ca