Louis 307

History TBA

Experiments

• April 21 2014

Start learning experiment - Mouse gets water reward all the time. Trial 1 : water always with GO trial - 10 min Trial 2 : 20 min only GO + Lick Trial 3 : 20 min only GO + Lick

• April 22 2014

Trial 1 -> Matlab crashed when saving behavior data (not video). Lost traces. Trial 2 : vertical bars Trial 3 : horizontal bars

• April 23 2014

30 min GO + No GO trials Went well 65 % Hit / 50 % Correct rejection

• April 24 2014

Went well, same at April 23

• April 25 2014

Made mistake with valve duration while setting up. Mouse got water accidently. Mouse was completely loosing attention. I clapped my hand at the beginning and it came back immediately. Should be visible in calcium.

• April 28 2014

Much better at task

• April 29 2014

Licking all the time (100% HITS / 95 % False Alarms)

• April 30 2014

Same as previous day

• May 1st 2014

Start to discriminate

• May 2nd 2014

Mouse good

• May 5th 2014

Perform well now

• May 6th 2014

Same

• May 7th 2014 Solve a lot of problems with timing in program. Move to Matlab for behavior movie recording to avoid drop frame. Behavior traces were unaffected. Improve timing of auditory cue and total duration of visual protocol. Do behavior next.

• May 8th 2014

Start constrast invariant experiment

• May 9th 2014

Same as day before.

Analysis

Classification By Oleg:

Building Naiive Bayes Calssifier.

Tried May07 Hit vs CR. ~85-90% correct performance of the calssifier. I perform subsampling in space(by 10) and time(by 5), then apply F-Fmin, then run PCA-ICA with low number of ICs(10-20) to extract 'blobs of activity'. Then threshold IC filters and reapply to the movie. This way traces for ICs between different Trials become stereotypical and without reapplying thresholded ICs it is a complete mess.

Training and test data are vectors of single trial duration for all ICs concatinated in one row.

Trying May 01 CR vs FA. In other days this ration is either too large or too small.