1.1 --- a/thesis/Makefile	Wed Mar 26 02:42:01 2014 -0400
     1.2 +++ b/thesis/Makefile	Wed Mar 26 03:18:57 2014 -0400
     1.3 @@ -1,6 +1,7 @@
     1.4  #INVOKE_LATEX = pdflatex -shell-escape thesis.tex;
     1.5  THESIS_NAME  = rlm-cortex-meng
     1.6  INVOKE_LATEX = texi2dvi --shell-escape --pdf -V --batch $(THESIS_NAME).tex;
     1.7 +#INVOKE_LATEX = texi2dvi --shell-escape --pdf -V  $(THESIS_NAME).tex;
     1.8  
     1.9  all:
    1.10  	./weave-thesis.sh cortex

     2.1 --- a/thesis/cortex.org	Wed Mar 26 02:42:01 2014 -0400
     2.2 +++ b/thesis/cortex.org	Wed Mar 26 03:18:57 2014 -0400
     2.3 @@ -663,8 +663,41 @@
     2.4  
     2.5  ** Empathy is the process of tracing though \Phi-space 
     2.6  
     2.7 +   Here is the core of a basic empathy algorithm, starting with an
     2.8 +   experience vector: First, group the experiences into tiered
     2.9 +   proprioceptive bins. I use powers of 10 and 3 bins, and the
    2.10 +   smallest bin has and approximate size of 0.001 radians in all
    2.11 +   proprioceptive dimensions.
    2.12 +   
    2.13 +   Then, given a sequence of proprioceptive input, generate a set of
    2.14 +   matching experience records for each input. 
    2.15  
    2.16 +   Finally, to infer sensory data, select the longest consective chain
    2.17 +   of experiences as determined by the indexes into the experience
    2.18 +   vector. 
    2.19  
    2.20 +   This algorithm has three advantages: 
    2.21 +
    2.22 +   1. It's simple
    2.23 +
    2.24 +   3. It's very fast -- both tracing through possibilites and
    2.25 +      retrieving possible interpretations take essentially constant
    2.26 +      time. 
    2.27 +
    2.28 +   2. It protects from wrong interpretations of transient ambiguous
    2.29 +      proprioceptive data : for example, if the worm is flat for just
    2.30 +      an instant, this flattness will not be interpreted as implying
    2.31 +      that the worm has its muscles relaxed, since the flattness is
    2.32 +      part of a longer chain which includes a distinct pattern of
    2.33 +      muscle activation. A memoryless statistical model such as a
    2.34 +      markov model that operates on individual frames may very well
    2.35 +      make this mistake.
    2.36 +
    2.37 +   #+caption: Program to convert an experience vector into a 
    2.38 +   #+caption: proprioceptively binned lookup function.
    2.39 +   #+name: bin
    2.40 +   #+begin_listing clojure
    2.41 +   #+begin_src clojure
    2.42  (defn bin [digits]
    2.43    (fn [angles]
    2.44      (->> angles
    2.45 @@ -674,11 +707,10 @@
    2.46           (mapv #(Math/round (* % (Math/pow 10 (dec digits))))))))
    2.47  
    2.48  (defn gen-phi-scan 
    2.49 -"Nearest-neighbors with spatial binning. Only returns a result if
    2.50 - the propriceptive data is within 10% of a previously recorded
    2.51 - result in all dimensions."
    2.52 -
    2.53 -[phi-space]
    2.54 +  "Nearest-neighbors with binning. Only returns a result if
    2.55 +   the propriceptive data is within 10% of a previously recorded
    2.56 +   result in all dimensions."
    2.57 +  [phi-space]
    2.58    (let [bin-keys (map bin [3 2 1])
    2.59          bin-maps
    2.60          (map (fn [bin-key]
    2.61 @@ -686,12 +718,20 @@
    2.62                  (comp bin-key :proprioception phi-space)
    2.63                  (range (count phi-space)))) bin-keys)
    2.64          lookups (map (fn [bin-key bin-map]
    2.65 -                      (fn [proprio] (bin-map (bin-key proprio))))
    2.66 -                    bin-keys bin-maps)]
    2.67 +                       (fn [proprio] (bin-map (bin-key proprio))))
    2.68 +                     bin-keys bin-maps)]
    2.69      (fn lookup [proprio-data]
    2.70        (set (some #(% proprio-data) lookups)))))
    2.71 +   #+end_src
    2.72 +   #+end_listing
    2.73  
    2.74  
    2.75 +   #+caption: Program to calculate empathy by tracing though \Phi-space
    2.76 +   #+caption: and finding the longest (ie. most coherent) interpretation
    2.77 +   #+caption: of the data.
    2.78 +   #+name: longest-thread
    2.79 +   #+begin_listing clojure
    2.80 +   #+begin_src clojure
    2.81  (defn longest-thread
    2.82    "Find the longest thread from phi-index-sets. The index sets should
    2.83     be ordered from most recent to least recent."
    2.84 @@ -718,6 +758,9 @@
    2.85                      threads)]
    2.86          (recur (concat longest-thread result)
    2.87                 (drop (count longest-thread) phi-index-sets))))))
    2.88 +   #+end_src
    2.89 +   #+end_listing
    2.90 +
    2.91  
    2.92  There is one final piece, which is to replace missing sensory data
    2.93  with a best-guess estimate. While I could fill in missing data by
    2.94 @@ -747,12 +790,14 @@
    2.95     #+end_src
    2.96     #+end_listing
    2.97  
    2.98 -
    2.99 -
   2.100 -
   2.101    
   2.102  ** Efficient action recognition with =EMPATH=
   2.103  
   2.104 +   In my exploration with the worm, I can generally infer actions from
   2.105 +   proprioceptive data exactly as well as when I have the complete
   2.106 +   sensory data. To reach this level, I have to train the worm with
   2.107 +   verious exercices for about 1 minute.
   2.108 +
   2.109  ** Digression: bootstrapping touch using free exploration
   2.110  
   2.111  * Contributions