Register now After registration you will be able to apply for this opportunity online.
This opportunity is not published. No applications will be accepted.
Machine Learning Evaluation of A Multi-User High-Performance Computer
Explore how machine learning algorithms are impacted by the several choices of general-purpose computing capable (not only graphics) of our GPU station to produce a high computation environment with minimal cost. More importantly, our system will be recommended to facilitate computational tasks of research projects for faculty members and students.
Keywords: Algorithm complexity, computational power, GPU performance, big data, high performance
The main goal of this thesis project is to develop a model called Dynamic task
resource allocation algorithm. This model takes as input machine learning
models/algorithms and the current hardware configuration of a high-performance cluster and outputs a policy for running the current task on the
cluster, enabling the training or deployment of as many machine learning
models and algorithms as possible on a multi-user cluster with as lower overhead as possible.
For multi-user high-performance clusters, this work is valuable. The computing cluster is abstracted as a platform to the user, the user usually is not aware of the specs of the cluster hardware and the user may not have the knowledge
to optimize training and deploy machine learning models on specific hardware. The goal of this work is that after a user submits a task, the specific
computational resources could be allocated according to the characteristics of the task itself (model features).
The main goal of this thesis project is to develop a model called Dynamic task resource allocation algorithm. This model takes as input machine learning models/algorithms and the current hardware configuration of a high-performance cluster and outputs a policy for running the current task on the cluster, enabling the training or deployment of as many machine learning models and algorithms as possible on a multi-user cluster with as lower overhead as possible. For multi-user high-performance clusters, this work is valuable. The computing cluster is abstracted as a platform to the user, the user usually is not aware of the specs of the cluster hardware and the user may not have the knowledge to optimize training and deploy machine learning models on specific hardware. The goal of this work is that after a user submits a task, the specific computational resources could be allocated according to the characteristics of the task itself (model features).
- Implement and run different machine learning algorithms (simple and complex) on different hardware configurations (CPU, CPU/GPU, combined).
- Data and mode: Serialization vs. parallelization.
- Resource allocation strategy and implementation
**Tasks**
- Literature review (5%)
- Implement and run machine learning algorithms sequentially and in parallel (30%)
- Design software for testing exaction performance (30%)
- Test, compare and evaluate results from CPU, one GPU, and two GPUs (30%)
- Report and present results (5%)
**Your Profile**
- Background in Computer Science, Software Engineering, Biostatistics, or related fields.
- Prior experience with programming (Matlab or Python).
- Experience with GPU computing is a plus.
- Able to work independently, pay attention to detail, and deliver results remotely.
- Can visualize data effectively using different charts such as boxplots and scatter plots.
- Background in statistics, time series analysis, and machine learning is needed.
- Implement and run different machine learning algorithms (simple and complex) on different hardware configurations (CPU, CPU/GPU, combined). - Data and mode: Serialization vs. parallelization. - Resource allocation strategy and implementation
**Tasks**
- Literature review (5%) - Implement and run machine learning algorithms sequentially and in parallel (30%) - Design software for testing exaction performance (30%) - Test, compare and evaluate results from CPU, one GPU, and two GPUs (30%) - Report and present results (5%)
**Your Profile**
- Background in Computer Science, Software Engineering, Biostatistics, or related fields. - Prior experience with programming (Matlab or Python). - Experience with GPU computing is a plus. - Able to work independently, pay attention to detail, and deliver results remotely. - Can visualize data effectively using different charts such as boxplots and scatter plots. - Background in statistics, time series analysis, and machine learning is needed.
Dr Moe Elgendi (moe.elgendi@hest.ethz.ch, Biomedical and Mobile Health Technology Research Group, ETH Zurich, http://bmht.hest.ethz.ch) will supervise the student during this project in collaboration with Dr. Jochen Klumpp (jochen.klumpp@hest.ethz.ch, ISG D-HEST, ETH Zurich).
Dr Moe Elgendi (moe.elgendi@hest.ethz.ch, Biomedical and Mobile Health Technology Research Group, ETH Zurich, http://bmht.hest.ethz.ch) will supervise the student during this project in collaboration with Dr. Jochen Klumpp (jochen.klumpp@hest.ethz.ch, ISG D-HEST, ETH Zurich).