Sara Shoouri, Morteza Tavakoli Taba Hun-Seok Kim
Proceedings of the AAAI Conference on Artificial Intelligence (AAAI), 2026.
Paper link
This paper proposes a predictive, history-aware adaptive scanning framework that anticipates informative regions of interest (ROI) based on past observations. Our approach introduces a lightweight predictor network that distills historical spatial and temporal contexts into refined query embeddings. These embeddings guide a differentiable Mask Generator network, which leverages Gumbel-Softmax sampling to produce binary masks identifying critical ROIs for the upcoming frame.
Qirui Zhang, Zichen Fan, Pierre Abillama Sara Shoouri, Jungho Lee, Chien-Wei Tseng, Wei Meng, Hun-Seok Kim, David Blaauw, Dennis Sylvester
Published in 2026 IEEE Journal of Solid-State Circuits
Paper link
This article presents MITTA, the first silicon-proven transformer accelerator optimized for multi-task inference across both natural language processing (NLP) and image processing domains. MITTA accelerates a task-sharing algorithm that minimizes sub-task computation by reusing both activations and weights from a shared base task, requiring only sparse delta computation for sub-tasks. To enhance efficiency, MITTA adopts block-wise structured sparsity (BSS) and INT8/INT4 mixed precision arithmetic, alongside a custom INT32 Softmax unit that avoids floating-point (FP) overhead while maintaining numerical accuracy and reducing power by 28%. To further optimize energy, MITTA implements a novel two-stage, task-adaptive dynamic voltage and frequency scaling (DVFS) scheme using a hierarchical power management architecture—combining off-chip DC–DC converters with on-chip power multiplexers (MUXes)—to dynamically select energy-optimal voltage–frequency operating points.
Zichen Fan, Qirui Zhang, Pierre Abillama Sara Shoouri, Jungho Lee, Chien-Wei Tseng, Wei Meng, Hun-Seok Kim, David Blaauw, Dennis Sylvester
Published in 2025 Symposium on VLSI Technology and Circuits (VLSI Technology and Circuits.)
Paper link
This paper presents the first silicon implementation of a transformer accelerator that executes multiple tasks simultaneously, improving per-task efficiency. It applies structured sparsity, multi-precision computation, and a two stage DVFS mechanism yielding highest reported efficiency per task.
Sara Shoouri, Mingyu Yang, Zichen Fan, Hun-Seok Kim
Accepted to ICCV, 2023
project page / arXiv / Presentation
We propose a novel framework that efficiently addresses multiple visual tasks by sharing knowledge through computation and parameter sharing among individually trained single-task transformers, outperforming SOTA in accuracy and resource utilization across image and video data.
Zichen Fan, Qirui Zhang, Pierre Abillama, Sara Shoouri, Changwoo Lee, David Blaauw, Hun-Seok Kim, Dennis Sylvester
Proceedings of the 50th Annual International Symposium on Computer Architecture (ISCA), 2023
Paper link
TaskFusion is an efficient software-hardware co-design for multi-task natural language processing (NLP), utilizing delta sparsity in weights and activations to reduce computation costs, achieving substantial performance and energy efficiency gains.
Morteza Tavakoli Taba, S. M. Hossein Naghavi, Sara Shoouri, Andreia Cathelin, Ehsan Afshari
Proceedings of IEEE 49th European Solid-State Circuits Conference (ESSCIRC), 2023
This paper introduces a novel technique for a 6-bit active differential phase shifter with accurate I/Q signals and minimal gain/phase errors.
Sara Shoouri, Mingyu Yang, Gordy Carichner, Yuyang Li, Ehab A Hamed, Angela Deng, Delbert A Green, Inhee Lee, David Blaauw, Hun-Seok Kim
IEEE Data Science and Learning Workshop (DSLW), 2022
project page / arXiv / Presentation
Proposed GPS-less method using deep learning sensor fusion improves daily Monarch butterfly tracking accuracy by utilizing daylight intensity and temperature data, achieving superior results compared to previous approaches.
Inhee Lee, Roger Hsiao, Gordy Carichner, Chin-Wei Hsu, Mingyu Yang, Sara Shoouri, Delbert A Green, Hun-Seok Kim, David Blaauw
Proceedings of the 27th Annual International Conference on Mobile Computing and Networking (MobiCom), 2021
Paper Link
MSAIL proposes a compact embedded system that harnesses solar energy and wireless transmission to track monarch butterfly migration, enabling daily location estimation through deep learning-based localization algorithms, validated in an outdoor experiment.
Sara Shoouri, Shayan Jalili, Jiahong Xu, Isabelle Gallagher, Yuhao Zhang, Joshua Wilhelm, Jean-Baptiste Jeannin, Necmiye Ozay
AIAA Scitech Forum, 2021
Presentation / Arxiv
We investigate camera-based automatic landing systems for smaller airports, proposing an architecture, validating specifications with flight data, and using the Breach tool to find counterexamples for falsification.
Yiran Shen, Reza Arablouei, Frank de Hoog, Jaques Malan, James Sharp, Sara Shoouri, Timothy D Clark, Carine Lefevre, Frederieke Kroon, Andrea Severati, Brano Kusy
ACM/IEEE International Conference on Information Processing in Sensor Networks, 2021
Paper Link
This study utilizes biologgers and a novel processing pipeline to detect fish feeding behavior via ECG signals, introducing an efficient change-detection algorithm for potential long-term monitoring of marine animal health in their natural habitats.
Yiran Shen, Reza Arablouei, Frank De Hoog, Xing Hao, Jacques Malan, James Sharp, Sara Shoouri, Timothy D Clark, Carine Lefevre, Frederieke Kroon, Andrea Severati, Branislav Kusy
IEEE Transactions on Mobile Computing, 2021
Paper Link
The study focuses on using implantable biologgers and a novel processing pipeline to detect feeding behavior in predatory fish via ECG signals, achieving accurate heart-rate estimation and feeding event detection.