Florian Maurer, M.Sc.

In latest research, machine learning has been successfully applied to many kinds of problems like picture classification or control. Machine learning has shown to be able to build highly accurate models also for a big amount of sensor data. In many problems, like autonomous driving, coordination between different entities (cars) which are controlled individually is necessary to optimize for a specific goal (routing with least average travelling time).

In such problems mainly two approaches exist: the (1) centralized and the (2) decentralized one. In the first (1) one, all sensor data is shared with a central controller which processes it. In the decentralized (2) manner only parts of the data is shared between the different entities, which decide on their actions locally based on the available information.

In our IPF project, we use rule-based RL (LCT) to optimize runtime parameters (DVFS, Task mapping) of MPSoCs.
Currently per-core LCTs optimize locally without knowledge about a common goal or the state of other LCTs and their cores.

This MA is investigating how to achieve coordination between the per-core LCTs by

• comparing known approaches in literature
• comparing thei applicability in simulation
• implementation and evaluation of one approach on an FPGA prototype

flo.maurer@tum.de

This seminar topic should result in a survey on state of the art scheduling methodes of tasks in NoCs and TSN connected ECU like in a car.

In contrast to the continued rise of various deep neural network ML variants, we favor the properties that come with explicit knowledge LCS (Learning Classifier System) -type ML techniques in our IPF project: interpretable and explainable rules which also can be pre-initialized with best-practice prior knowledge content.

This Seminar should provide an overview of existing, explainable ML techniques and of methodes to make black-box techniques (NNs) explainable.

The goal of this thesis is to improve the current task migration mechanism in our IPF platform. This one should be especially adapted to heterogeneous MPSoCs.

Therefore, different scheduling methods schould be investigated based on a literature study.

Before implementing in HW, simulations will be done in Matlab.

Today's Multi-Processor System-on-Chip (MPSoCs) are getting more and more complex due to the growing amount of cores and accelerators.  Hence it's not possible anymore to set runtime parameters like frequency and task distribution by design time in an optimal manner. Therefore future controllers try to make use of machine learning which is aware of the system's current state (self-awareness).

Information Processing Factory (IPF) is a global project that claims to show self-awareness across multiple abstraction levels. It represents a paradigm shift in platform design by envisioning the move towards a consistent platform-centric design in which the combination of self-organized learning and formal reactive methods guarantee the applicability of such cyber-physical systems in safety-critical and high-availability applications.

At TUM, we explore the application and implementation of machine learning algorithms in hardware to optimize the mode of operation of MPSoCs at runtime. 

Currently we are running a quite simple programm on our MPSoC which doesn't allow us to evaluate all the features our self-aware system provides.

Therefore, it would be your task to develop a easely configurable software to be executed on the MPSoC which schedules different benchmarks.

Towards this goal, you'll complete the following tasks:

1. Literature research on embedded benchmarks + scheduling mechanisms
2. Getting familiar with our SparcV8 (Leon 3) architecture
3. Porting different benchmarks
4. Implementing and evaluating your scheduler

• Good Knowledge about MPSoCs
• Good C knowledge
• Maybe bash and xconfig knowledge

flo.maurer@tum.de

Today's Multi-Processor System-on-Chip (MPSoCs) are getting more and more complex due to the growing amount of cores and accelerators.  Hence it's not possible anymore to set runtime parameters like frequency and task distribution by design time in an optimal manner. Therefore future controllers try to make use of machine learning which is aware of the system's current state (self-awareness).

Information Processing Factory (IPF) is a global project that claims to show self-awareness across multiple abstraction levels. It represents a paradigm shift in platform design by envisioning the move towards a consistent platform-centric design in which the combination of self-organized learning and formal reactive methods guarantee the applicability of such cyber-physical systems in safety-critical and high-availability applications.

At TUM, we explore the application and implementation of machine learning algorithms in hardware to optimize the mode of operation of MPSoCs at runtime. 

Therefore, we currently evaluate our research primary on FPGAs, but also by Gem5 simulations. Unfortunatelly, the FPGA evaluation causes long and quite complex developement and Gem5 simulation is time intensive when running experiments several times for different parameters or approaches.

To overcome these two issues, you will extend a simulation environment which performs trace-based DVFS (should be extended to Task Migration) simulation in Matlab towards a multi-core system.

Your environment should keep using Gem5 traces and provide performance and power metrics, as well as the current state like utilization as outputs when performing DVFS in a multicore system. Hereby, your environment has to account for shared resources.

Towards this goal, you'll complete the following tasks:

1. Literature research on the state of the art of trace-based simulation
2. Getting familiar with object orientated programming in Matlab
3. Developping your simulation environement in Matlab
4. Verifying your environement by comparing to Gem5

This topic can be extended to a Master Thesis.

• Good Knowledge about MPSoCs
• Good Matlab Knowledge
• Good Bash Knowledge
• Basic Knowledge about Trace-Based Simulation
• Maybe Gem5 Knowledge

flo.maurer@tum.de

Student projects will be continued and started also under the current circumstances (SARS-CoV-2). This task can be executed remotly by accessing our chair's computers, or a students personal device, as all needed programs are either free or a student license is available.