Announcements

From July 2024, new funds from Korea government (July. 2024~Dec. 2028) titled as Development and Operation of a Research and Education Program for Low-cost AI System Integration for Edge will begin, where HW, SW, and applications for Edge AI will be educated and researched with pronoun companies and graduate students. (4.5년 정부 11.25억 지원, 제목: Edge용 초저비용 AI 시스템 융합인재연구교육과정 개발 및 운영, 사업명: 디지털기술선도핵심인재양성(R&D), 세부사업명: 학석사연계ICT 핵심인재양성) 관련기사: https://www.hankyung.com/article/202407117168h



We start to manage our lab's organization GitHub repositories. Including last open source from our lab, new code and dataset will be released under the control of our lab. 


A domestic patent from the collaboration with SK Hynix titled as Similarity-based wafer failure pattern analysis method and apparatus using image matching(이미지 매칭을 이용한 유사도 기반 웨이퍼 고장 패턴 분석 방법 및 장치) has been submitted. Thanks for Young Wook Kwon and Soo min Oh. 


Three researchers of EmPasLab have participated in ICASSP 2024, Coex, South Korea. In a poster session, our paper was well presented in front of about 20 international researchers. You can see the picture of ICASSP 2024 (➡️pictures). 


I am trilled that our paper titled "A novel method to analysis the wafer defect patterns using an image matching algorithm based on deep neural networks" has been accepted in the Work-in-Progress (WIP) category at the 61th Design and Automation (DAC) conference, taking place on June 23-27th, 2024 at Moscone Center West in San Francisco, California. Although the paper is not selected into regular research papers, the result of the first year's research with SK Hynix are satisfactory. Better results can be expected later. Thanks for my advised students, Youngwook Kwon and Sumin Oh!! 


Fortunately, one AI clould machine (>20 TFLOPS) will be supported from NIPA this year. 


A domestic patent titled as "SYSTEMS FOR IMPLEMENTING BINARY NEURAL NETWORKS SUITABLE FOR PYRAMID STRUCTURE (피라미드 구조에 적합한 이진 뉴럴 네트워크 구현 시스템)" by HyunJin Kim has been registered. (Patent Number: 10-2632473)


A paper titled as "Low-Biased Probabilistic Multipliers using Complementary Inaccurate Compressors" by HyunJin Kim and Alberto A. Del Barrio has been accepted in Digital Signal Processing. This paper proposes new probabilistic multipliers using complementary inaccurate compressors for achieving better error characteristics. The hardware cost analysis and application for neural networks were also presented in this paper. For this acceptance, there are long journey. This new can be a good gift for this Christmas vacation.


Jisoo Shin and Yongbean Park will join a new M.S. candidate and an undergraduate researcher.  They will start their research from this winter vacation. Welcome sincerely!!


A paper titled as "1-D SPATIAL ATTENTION IN BINARIZED CONVOLUTIONAL NEURAL NETWORKS" by HyunJin Kim, Jungwoo Shin, Wansoo Kim, Alberto A. Del Barrio has been accepted in ICASSP (IEEE International Conference on Acoustics, Speech and Signal Processing) 2024. This is the flagship conference hosted by IEEE Signal Processing Society. About 6,000 papers were submitted to this conference, and it has 45% acceptance rate. 


From Nov. 1-3, 2023, several student researchers and Prof. Kim participated in K-Chips Workshop in Jeju. Along with presentation and educational courses, we enjoyed beautiful weather and remarkable ocean views. Besides, delicious foods were very good. 


From July 2023, new funds from Korea government (July. 2023~Dec. 2025) begin, where methods and user interface for building supervised data in wafer testing will be developed. (민관공동투자 반도체 고급인력 양성사업 주관기관 선정)


Our all members attended CKAIA2023 at Yeosu, South Korea and presented a poster titled as “ResidualViT: Better general-purpose Vision Transformer combined with CNN.” Notably, we can have good chance to understand LLM and generative AI. Besides, beautiful scenery of conference site and good dishes were another good things. 


Our members won the Best Paper Award at the 24th Korea Test Conference for their paper titled as Study on Wafer Failure Pattern Detection using Image Matching Algorithms. Congratulation.

 

Five researchers attended the 9-th Open Robotics Seminar, held at Robotis Inc., Seoul. There were 13 sessions for introducing ROS applications. Picture


Eight researchers enjoyed KCS 2023, held High One Resort, Kangwon-do. There were interesting topics including Processing-In-Memory, Hardware Accelerators, etc. 


We have five servers for machine learning and semiconductor design. These servers will be moved to our department's specific server room. We are happy to avoid noisy sound and hot air from them. 


Se-Jin Kwon, an undergraduate student, joined our EmPasLab. She has interest in digging the details and secrets of machine learning. 


At this time, Prof. Kim and Jung-Woo Shin have submitted their BNN (binarized neural networks) works to ICML 2023. ICML is one of top-tier AI conferences. Let's cross finger. 


During this winter vacation, several ideas and noticeable subjects for artificial intelligence and machine learning will be researched by student researchers and Prof. Kim. Notably, visual transformer, reinforcement learning for handling robot arm, TinyML, RISC-V based Posit arithmetic, BNNs are being focused. Good Luck!!


Young-Wook Kwon, a master candidate, joined our EmPasLab. He has interest in developing new visual transformer and inventing ideas for TinyML. 


From December 2022, new funds from SK Hynix (Dec. 2022~Nov. 2024) begin, where methods and user interface for building supervised data in wafer testing will be developed. 


At this time, advised students and Prof. Kim have submitted their TinyML works to ICCAD TinyML Contests. In a small embedded board, binarized neural networks can be implemented, which significantly reduces hardware costs in terms of latency and parameter storage. In this work, this team can fully understand practical implementation of TinyML and methods about how to debug TinyML results. 


At this time, Prof. Kim and other advised students have studied binarized neural networks (BNNs) from this summer. For 1-D data (e.g. sound), BNNs are a very realistic form to be implemented on lightweight MCUs. Besides, their low latency and low power consumption can extend the applicable area of neural networks. 


On June 26, 2022, a manuscript titled as "CTMQ: Cyclic Training of Convolutional Neural Networks with Multiple Quantization Steps" has been uploaded in arXiv.


On May 23/24, 2022, there is an open lab ceremony to explain the life of graduate schools and method for achieving research contributions. (#403, 2nd Engineering building, Dankook University) 


Prof. HyunJin Kim had a meeting with ARM to apply the ARM Academic Access (AAA) Program. The program supports ARM commercial IPs, training Program, and Materials for researches with ARM IPs.  


A paper titled "A storage-efficient ensemble classification using filter sharing on binarized convolutional neural networks" by HyunJin Kim, Mohammed Alnemari, and Nader Bagherzadeh has been accepted in PeerJ Computer Science.


A paper titled "PresB-Net: parametric binarized neural network with learnable activations and shuffled grouped convolution" by Jungwoo Shin and HyunJin Kim has been accepted in PeerJ Computer Science. 


A paper titled "Highly accurate approximate multiplier using heterogeneous inexact 4-2 compressors for error-resilient application" by Jaewoo Lee and HyunJin Kim has been accepted in IEMEK Journal of embedded Systems and Applications (Domestic). 


A paper titled "A Cost-Efficient Approximate Dynamic Ranged Multiplication and Approximation-Aware Training on Convolutional Neural Networks" by HyunJin Kim and Alberto A. Del Barrio has been accepted in IEEE Access. 


A paper titled "PLAM: a Posit Logarithm-Approximate Multiplier" by Raul Murillo, Alberto A. Del Barrio, Guillermo Botella, Min Soo Kim, HyunJin Kim, Nader Bagherzadeh has been accepted in IEEE Transactions on Emerging Topics in Computing. 


A project titled  "ACDNN: Approximate Computing-based Deep Neural Networks using Inaccurate Arithmetic Units for Low-Power Systems" will be supported by National Research Foundation (NRF) (June 2021 ~ February 2024) - 2021.05.2


A paper titledA k-Mismatch String Matching for Generalized Edit Distance using Diagonal Skipping Method” by HyunJin Kim has been accepted in PLOS One.


A paper titled "AresB-Net: accurate residual binarized neural networks using shortcut concatenation and shuffled grouped convolution" by HyunJin Kim has been accepted in PeerJ Computer Science


A paper titled "A Low-Cost Compensated Approximate Multiplier for Bfloat16 Data Processing on CNN Inference" by HyunJin Kim has been accepted in ETRI Journal


A paper titled "Effects of Approximate Multiplication on Convolutional Neural Networks," Kim, M. S., Del Barrio, A. A., Kim, H., & Bagherzadeh, N has been published in the early access of IEEE Transactions on Emerging Topics in Computing


A domestic patent titled "System and Method for Preprocessing and Data Set Augmentation for training AIwith 3D Data Processing" is decided to be registered. 


A Paper titled "Analysis of Reduced-Width Truncated Mitchell Multiplication for Inferences Using CNNs" has been published in IEMEK Journal. 



As the result of collaboration with UCI researchers, Prof. HyunJin Kim will present his work in 26-th IEEE Symposium on Computer Arithmetic (ARITH). ARITH 2019 will be held in Kyoto, Japan. This paper is titled as "A Cost-Efficient Iterative Truncated Logarithmic Multiplication for Convolutional Neural Networks." This research will be extended with additional error analysis and full ImageNet database. 


As the result of collaboration with UCI researchers, Prof. HyunJin Kim will present a poster in Design Automation Conference 2019. DAC 2019 will be held in Las Vegas, NV, USA. This poster is titled as "A Cost Effective, High-Accuracy Two-Stage Approximate Logarithmic Multiplier." This research will be extended with the experiments applied to neural networks.