| 1. |
- Wärnberg, Emil
(författare)
-
On learning in mice and machines : continuous population codes in natural and artificial neural networks
- 2023
-
Doktorsavhandling (övrigt vetenskapligt/konstnärligt)abstract
- Neural networks, whether artificial in a computer or natural in the brain, could represent information either using discrete symbols or continuous vector spaces. In this thesis, I explore how neural networks can represent continuous vector spaces, using both simulated neural networks and analysis of real neural population data recorded from mice. A special focus is on the networks of the basal ganglia circuit and on reinforcement learning, i.e., learning from rewards and punishments.The thesis includes four scientific papers: two theoretical/computational (Papers I and IV) and two with analysis of real data (Papers II and III).In Paper I, we explore methods for implementing continuous vector spaces in networks of spiking neurons using multidimensional attractors, and propose an explanation for why it is hard to escape the neural manifolds created by such attractors.In Paper II, we analyze experimental data from dorsomedial striatum collected using 1-photon calcium imaging of transgenic mice with celltype-specific markers for the striatal direct, indirect and patch pathways, as the mice were gathering rewards in a 2-choice task. In line with extensive previous results, our data analysis revealed a number of neural signatures of reinforcement learning, but no apparent difference between the pathways.In Paper III, we present a new software tool for tracking neurons across weeks of 1-photon calcium imaging, and employ it to follow patch-specific striatal projection neurons from the dorsomedial striatum across two weeks of daily recordings.In Paper IV, we propose a model for how the nigrostriatal dopaminergic projection could, in a biologically plausible way, convey a vector-valued error gradient to the dorsal striatum, as required for backpropagation.Based on the results of the papers and a review of existing literature, I argue that while the basal ganglia indeed make up a circuit for reinforcement learning as previously thought, this circuit represents reinforcement learning states, actions and policies using a continuous population code and not using discrete symbols.
|
|
| 2. |
- Belic, Jovana
(författare)
-
Untangling Cortico-Striatal Circuitry and its Role in Health and Disease - A computational investigation
- 2018
-
Doktorsavhandling (övrigt vetenskapligt/konstnärligt)abstract
- The basal ganglia (BG) play a critical role in a variety of regular motor and cognitive functions. Many brain diseases, such as Parkinson’s diseases, Huntington’s disease and dyskinesia, are directly related to malfunctions of the BG nuclei. One of those nuclei, the input nucleus called the striatum, is heavily connected to the cortex and receives afferents from nearly all cortical areas. The striatum is a recurrent inhibitory network that contains several distinct cell types. About 95% of neurons in the striatum are medium spiny neurons (MSNs) that form the only output from the striatum. Two of the most examined sources of GABAergic inhibition into MSNs are the feedback inhibition (FB) from the axon collaterals of the MSNs themselves, and the feedforward inhibition (FF) via the small population (1-2% of striatal neurons) of fast spiking interneurons (FSIs). The cortex sends direct projections to the striatum, while the striatum can affect the cortex only indirectly through other BG nuclei and the thalamus. Understanding how different components of the striatal network interact with each other and influence the striatal response to cortical inputs has crucial importance for clarifying the overall functions and dysfunctions of the BG. In this thesis I have employed advanced experimental data analysis techniques as well as computational modelling, to study the complex nature of cortico-striatal interactions. I found that for pathological states, such as Parkinson’s disease and L-DOPA-induced dyskinesia, effective connectivity is bidirectional with an accent on the striatal influence on the cortex. Interestingly, in the case of L-DOPA-induced dyskinesia, there was a high increase in effective connectivity at ~80 Hz and the results also showed a large relative decrease in the modulation of the local field potential amplitude (recorded in the primary motor cortex and sensorimotor striatum in awake, freely behaving, 6-OHDA lesioned hemi-parkinsonian rats) at ~80 Hz by the phase of low frequency oscillations. These results suggest a lack of coupling between the low frequency activity of a presumably larger neuronal population and the synchronized activity of a presumably smaller group of neurons active at 80 Hz. Next, I used a spiking neuron network model of the striatum to isolate the mechanisms underlying the transmission of cortical oscillations to the MSN population. I showed that FSIs play a crucial role in efficient propagation of cortical oscillations to the MSNs that did not receive direct cortical oscillations. Further, I have identified multiple factors such as the number of activated neurons, ongoing activity, connectivity, and synchronicity of inputs that influenced the transfer of oscillations by modifying the levels of FB and FF inhibitions. Overall, these findings reveal a new role of FSIs in modulating the transfer of information from the cortex to striatum. By modulating the activity and properties of the FSIs, striatal oscillations can be controlled very efficiently. Finally, I explored the interactions in the striatal network with different oscillation frequencies and showed that the features of those oscillations, such as amplitude and frequency fluctuations, can be influenced by a change in the input intensities into MSNs and FSIs and that these fluctuations are also highly dependent on the selected frequencies in addition to the phase offset between different cortical inputs. Lastly, I investigated how the striatum responds to cortical neuronal avalanches. Recordings in the striatum revealed that striatal activity was also characterized by spatiotemporal clusters that followed a power law distribution albeit, with significantly steeper slope. In this study, an abstract computational model was developed to elucidate the influence of intrastriatal inhibition and cortico-striatal interplay as important factors to understand the experimental findings. I showed that one particularly high activation threshold of striatal nodes can reproduce a power law-like distribution with a coefficient similar to the one found experimentally. By changing the ratio of excitation and inhibition in the cortical model, I saw that increased activity in the cortex strongly influenced striatal dynamics, which was reflected in a less negative slope of cluster size distributions in the striatum. Finally, when inhibition was added to the model, cluster size distributions had a prominently earlier deviation from the power law distribution compared to the case when inhibition was not present.
|
|
