PRObablistic on-edge learning for SPINtronic-based neuromorphic systems
The hired joint UGA – KIT PHD candidate should be able to cover the work of the workpackage 1 and 2. He/she will also participate to technical meetings and have a good understanding on how the tasks of the other technical workpackages are executed, mainly by the partners with internal effort. As a whole, the PHD candidate will develop and optimize compact Computing in Memory architectures, provide high level models for further integration in large scale designs, perform validation of all proofs of concepts of new architectural implementations. He/she will be involved also in the design of algorithmic implementations of Bayesian Neural Networks adapted to the architecture. More in details, he/she will work on the following directions:
Design and optimization of the probabilistic neural networks, will be executed mostly in SPINTEC Laboratory in Grenoble, that will include:
1. full design stack of hardware accelerator without selector transistor for frequent Read and Write operations.
2. Design and validate an innovative architectural approach able to compensate for sneaky paths phenomena.
3. High-level modeling of the full crossbar architecture that includes the stochastic component.
4. Propose a full simulation and validation flow scalable to scaled to realistic architecture size and parameters that implement Bayesian tasks.
5. Perform Delay, power consumption and area overhead figures of merit
Study of the links between the dysregulations of metabolism and epigenetics marks in Huntington’s disease
We want to focus on epigenetic dysregulation in Huntington’s Disease (HD), a pathogenic mechanism implicated in accelerated aging of striatal neurons. Specifically, we will investigate the interplay between altered energy metabolism and epigenetic impairment in HD striatal neurons to identify new targets/pathways for disease-modifying intervention. We aim to obtain detailed maps of histone post-translational modifications (PTMs), especially of methylations, acetylation and the recently described lactylation, which might be critical in the HD brain. Indeed, these PTMs are tightly regulated by the metabolic status of the cells. We will use proteomics which is the best suited approach to identify and quantify multiple protein PTMs. We consider working on the striatum of WT, R6/1 transgenic mice and the more progressive Q140 knock in model at various stages of disease, to assess evolution of histone PTMs and metabolism with aging. Additionally, to get a dynamic view of the links between metabolic and epigenetic imbalance in HD, we will inject intraperitoneally HD mice and controls with 13C-glucose and analyze over a time course the incorporation of 13C into histone PTMs. Finally, acetyl-CoA, the precursor for histone lysine acetylation, has been shown to be locally produced in the nucleus, by either acetyl-CoA synthetase 2 (ACSS2), ATP-citrate lyase (ACLY) or the pyruvate dehydrogenase complex. Regarding lactylation, it is currently unknown where, and by which enzymes, the pool of lactate used for modifying histone lysines by lactylation is produced. ACSS2 is a very good candidate, as it can catalyze the production of acyl-CoA molecules from the corresponding short chain fatty acids (SCFA). To address the question of the production of metabolites in the vicinity of chromatin in striatal cells, we will use epigenomics (ChIPseq or CUT&tag) to get the genomic distribution of ACSS2 and ACLY and compare it to distributions of acetyl and lactyl histone marks.