Academia Sinica Logo

Research on Optimization of Deep Learning Model Inference and Training

The Computer Systems Laboratory - Machine Learning Systems team focuses on research areas including parallel and distributed computing, compilers, and computer architecture. We aim to leverage computer system technologies to accelerate the inference and training of deep learning models and develop optimizations for next-generation AI models. Our research emphasizes the following:

1. AI Model Compression and Optimization
Model compression techniques (e.g., pruning and quantization) reduce the size and computational demands of AI models, which are crucial for resource-constrained platforms such as embedded systems and memory-limited AI accelerators. We aim to explore:
* AI compiler: deployment methods for compressed models across servers, edge devices, and heterogeneous systems.
* High performance computing: efficient execution of compressed models on processors with advanced AI extensions, e.g., Intel AVX512, ARM SVE, RISC-V RVV, and tensor-level accelerations on GPUs and NPUs.

2. AI Accelerator Design
We aim to design AI accelerators for accelerating AI model inference, focusing on software and hardware co-design and co-optimization.

3. Optimization of AI Model Inference in Heterogeneous Environments
Computer architectures are evolving toward heterogeneous multi-processor designs (e.g., CPUs + GPUs + AI accelerators). Integrating heterogeneous processors to execute complex models (e.g., hybrid models, multi-models, and multi-task models) with high computational efficiency poses a critical challenge. We aim to explore:
* Efficient scheduling algorithms.
* Parallel algorithms for the three dimensions: data parallelism, model parallelism, and tensor parallelism.

We are looking to hire experienced mobile developers to help bring the PQConnect project to Android and iOS. PQConnect is a new easy-to-install layer of Internet security. It automatically applies post-quantum cryptography from end to end between computers running PQConnect, adding cryptographic protection to unencrypted applications, working in concert with existing pre-quantum applications to add post-quantum protection, and adding a second application-independent layer of defense to any applications that have begun to incorporate application-specific post-quantum protection. PQConnect works at the network layer similarly to a VPN, but is scalable in ways that most VPN applications are not. You can see more information at https://www.pqconnect.net

Our research team is seeking a Postdoctoral Researcher to join us in studying the ecology and evolution of microbial symbiosis in plant–microbe systems. We integrate field ecology, molecular microbiology, and genomic analyses to understand how microbial communities assemble and adapt across hosts and environments. This position is open until filled.
Current research topics
- The effect of host identity on symbiont community assembly
- Ecological genomics of symbionts across co-occuring native plant hosts
- Functional variation in ecologically important traits in faculative symbionts

Job description
- Lead and particiapte in research projects
- Design and conduct experiments and data analyses
- Actively contribute to scientific writing and presentations
- Mentor junior researchers and assist in training activities
- Manage routine laboratory tasks and other assigned duties

• 從事腦機介面或AR/VR/MR的相關研究。
• 提出假設，設計實驗，並進行數據收集。
• 分析電生理訊號（包括但不限於EEG、EMG、EOG、SSVEP、眼球追蹤器等）並彙總發現。
• 撰寫科學論文並發表成果。
• 與其他團隊成員合作，記錄和轉移開發的技術。
• Carry out research in the field of Brain-Computer Interfaces or AR/VR/MR.
• Formulate hypotheses, plan user studies, and carry out data collection.
• Analyze the collected bio-signals (include but not limited to EEG, EMG, EOG, SSVEP, eye-tracker.. ) and summarize the findings.
• Contribute to the design of innovative devices for collecting bio-signals and test them.
• Write scientific papers detailing findings for scientific conferences and journals.
• Cooperate with other team members and business groups on documenting and transferring the developed technologies.

All position.
To conduct an innovative research program in computational (broadly-defined), materials, energy, catalysis, chemical biology or other chemistry areas.

【Research Focus】
The Lab for Cell Dynamics (LCD) at Academia Sinica explores the biological functions of large-scale cell death during embryonic development, integrating advanced quantitative and imaging techniques. Our recent work (PMID: 38987590) reveals how developmental signals prime cells for death and how this process might influence aging. We invite passionate junior scientists to join our team to investigate:
1.How developmental cues orchestrate cell death during development.
2.The potential link between developmental death signaling and the aging process.

Conduct methodological and collaborative research in causal inference, survival analyses and statistical genomics with applications in studies of cancer and chronic diseases.