1.1 --- a/thesis/cortex.org	Wed Mar 26 02:42:01 2014 -0400
     1.2 +++ b/thesis/cortex.org	Wed Mar 26 03:18:57 2014 -0400
     1.3 @@ -663,8 +663,41 @@
     1.4  
     1.5  ** Empathy is the process of tracing though \Phi-space 
     1.6  
     1.7 +   Here is the core of a basic empathy algorithm, starting with an
     1.8 +   experience vector: First, group the experiences into tiered
     1.9 +   proprioceptive bins. I use powers of 10 and 3 bins, and the
    1.10 +   smallest bin has and approximate size of 0.001 radians in all
    1.11 +   proprioceptive dimensions.
    1.12 +   
    1.13 +   Then, given a sequence of proprioceptive input, generate a set of
    1.14 +   matching experience records for each input. 
    1.15  
    1.16 +   Finally, to infer sensory data, select the longest consective chain
    1.17 +   of experiences as determined by the indexes into the experience
    1.18 +   vector. 
    1.19  
    1.20 +   This algorithm has three advantages: 
    1.21 +
    1.22 +   1. It's simple
    1.23 +
    1.24 +   3. It's very fast -- both tracing through possibilites and
    1.25 +      retrieving possible interpretations take essentially constant
    1.26 +      time. 
    1.27 +
    1.28 +   2. It protects from wrong interpretations of transient ambiguous
    1.29 +      proprioceptive data : for example, if the worm is flat for just
    1.30 +      an instant, this flattness will not be interpreted as implying
    1.31 +      that the worm has its muscles relaxed, since the flattness is
    1.32 +      part of a longer chain which includes a distinct pattern of
    1.33 +      muscle activation. A memoryless statistical model such as a
    1.34 +      markov model that operates on individual frames may very well
    1.35 +      make this mistake.
    1.36 +
    1.37 +   #+caption: Program to convert an experience vector into a 
    1.38 +   #+caption: proprioceptively binned lookup function.
    1.39 +   #+name: bin
    1.40 +   #+begin_listing clojure
    1.41 +   #+begin_src clojure
    1.42  (defn bin [digits]
    1.43    (fn [angles]
    1.44      (->> angles
    1.45 @@ -674,11 +707,10 @@
    1.46           (mapv #(Math/round (* % (Math/pow 10 (dec digits))))))))
    1.47  
    1.48  (defn gen-phi-scan 
    1.49 -"Nearest-neighbors with spatial binning. Only returns a result if
    1.50 - the propriceptive data is within 10% of a previously recorded
    1.51 - result in all dimensions."
    1.52 -
    1.53 -[phi-space]
    1.54 +  "Nearest-neighbors with binning. Only returns a result if
    1.55 +   the propriceptive data is within 10% of a previously recorded
    1.56 +   result in all dimensions."
    1.57 +  [phi-space]
    1.58    (let [bin-keys (map bin [3 2 1])
    1.59          bin-maps
    1.60          (map (fn [bin-key]
    1.61 @@ -686,12 +718,20 @@
    1.62                  (comp bin-key :proprioception phi-space)
    1.63                  (range (count phi-space)))) bin-keys)
    1.64          lookups (map (fn [bin-key bin-map]
    1.65 -                      (fn [proprio] (bin-map (bin-key proprio))))
    1.66 -                    bin-keys bin-maps)]
    1.67 +                       (fn [proprio] (bin-map (bin-key proprio))))
    1.68 +                     bin-keys bin-maps)]
    1.69      (fn lookup [proprio-data]
    1.70        (set (some #(% proprio-data) lookups)))))
    1.71 +   #+end_src
    1.72 +   #+end_listing
    1.73  
    1.74  
    1.75 +   #+caption: Program to calculate empathy by tracing though \Phi-space
    1.76 +   #+caption: and finding the longest (ie. most coherent) interpretation
    1.77 +   #+caption: of the data.
    1.78 +   #+name: longest-thread
    1.79 +   #+begin_listing clojure
    1.80 +   #+begin_src clojure
    1.81  (defn longest-thread
    1.82    "Find the longest thread from phi-index-sets. The index sets should
    1.83     be ordered from most recent to least recent."
    1.84 @@ -718,6 +758,9 @@
    1.85                      threads)]
    1.86          (recur (concat longest-thread result)
    1.87                 (drop (count longest-thread) phi-index-sets))))))
    1.88 +   #+end_src
    1.89 +   #+end_listing
    1.90 +
    1.91  
    1.92  There is one final piece, which is to replace missing sensory data
    1.93  with a best-guess estimate. While I could fill in missing data by
    1.94 @@ -747,12 +790,14 @@
    1.95     #+end_src
    1.96     #+end_listing
    1.97  
    1.98 -
    1.99 -
   1.100 -
   1.101    
   1.102  ** Efficient action recognition with =EMPATH=
   1.103  
   1.104 +   In my exploration with the worm, I can generally infer actions from
   1.105 +   proprioceptive data exactly as well as when I have the complete
   1.106 +   sensory data. To reach this level, I have to train the worm with
   1.107 +   verious exercices for about 1 minute.
   1.108 +
   1.109  ** Digression: bootstrapping touch using free exploration
   1.110  
   1.111  * Contributions