dodox/Spikey
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

Better nomenclauture for README

Browse Source
This commit is contained in:
Dominik Moritz Roth 2024-05-27 17:08:11 +02:00
parent 8bf85d1a65
commit 973c5c89ba
1 changed files with 3 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

6
README.md
View File

@ -12,13 +12,13 @@ The Neuralink N1 implant generates approximately 200 Mbps of electrode data (102
## Data Analysis
The `analysis.ipynb` notebook contains a detailed analysis of the data. We found that there is sometimes significant cross-correlation between the different leads, so we find it vital to use this information for better compression. This cross-correlation allows us to improve the accuracy of our predictions and reduce the overall amount of data that needs to be transmitted.
The `analysis.ipynb` notebook contains a detailed analysis of the data. We found that there is sometimes significant cross-correlation between the different threads, so we find it vital to use this information for better compression. This cross-correlation allows us to improve the accuracy of our predictions and reduce the overall amount of data that needs to be transmitted.
## Algorithm Overview
### 1 - Thread Topology Reconstruction
As the first step, we analyze readings from the leads to construct an approximate topology of the threads in the brain. The distance metric we generate only approximately represents true Euclidean distances, but rather the 'distance' in common activity. This topology must only be computed once for a given implant and may be updated for thread movements but is not part of the regular compression/decompression process.
As the first step, we analyze readings from the threads to construct an approximate topology of the threads in the brain. The distance metric we generate only approximately represents true Euclidean distances, but rather the 'distance' in common activity. This topology must only be computed once for a given implant and may be updated for thread movements but is not part of the regular compression/decompression process.
### 2 - Predictive Model
@ -30,7 +30,7 @@ We separate the predictive model into four parts:
2. **Latent Projector**: This takes the feature vectors and projects them into a latent space. The latent projector can be configured as a fully connected network or an RNN (LSTM) with an arbitrary shape.
3. **[MiddleOut](https://www.youtube.com/watch?v=l49MHwooaVQ)**: For each lead, this module performs message passing according to the thread topology. Their latent representations along with their distance metrics are used to generate region latent representations. This is done by training a fully connected layer to map from (our_latent, their_latent, metric) -> region_latent and then averaging over all region_latent values to get the final representation.
3. **[MiddleOut](https://www.youtube.com/watch?v=l49MHwooaVQ)**: For each thread, this module performs message passing according to the thread topology. Their latent representations along with their distance metrics are used to generate region latent representations. This is done by training a fully connected layer to map from (our_latent, their_latent, metric) -> region_latent and then averaging over all region_latent values to get the final representation.
4. **Predictor**: This module takes the new latent representation from the MiddleOut module and predicts the next timestep. The goal is to minimize the prediction error during training. It can be configured to be an FCNN of arbitrary shape.