Minor changes <noupdate>

src/year1/cognition-and-neuroscience/module1/sections/_pavlovian_learning.tex

@@ -77,13 +77,14 @@ There are two types of learning:
 
     The extinct association can return in the future
     (this is more evident when the context is the same as the acquisition phase).
+
+    \begin{figure}[H]
+        \centering
+        \includegraphics[width=0.95\linewidth]{./img/pavlovian_extinction.png}
+        \caption{Example of acquisition, extinction, and \ac{cr} return}
+    \end{figure}
 \end{remark}
 
-\begin{figure}[H]
-    \centering
-    \includegraphics[width=0.95\linewidth]{./img/pavlovian_extinction.png}
-    \caption{Example of acquisition, extinction, and \ac{cr} return}
-\end{figure}
 
 \begin{description}
     \item[Generalization] \marginnote{Generalization} 
@@ -469,7 +470,7 @@ There is strong evidence that the dopaminergic system is the major neural mechan
         \end{casestudy}
 
     \item[Probability encoding] \marginnote{Dopamine probability encoding}
-        The phasic activation of dopamine neurons varies monotonically with the reward probability
+        The phasic activation of dopamine neurons varies monotonically with the reward probability.
         \begin{casestudy}
             \phantom{}
             \begin{center}

src/year1/cognition-and-neuroscience/module1/sections/_rl.tex

@@ -22,6 +22,7 @@
 \begin{remark}
     Multiple competing sub-systems contribute to learning and controlling behavior in animals.
 
+    \indenttbox
     \begin{example}[Freud's theory of the mind structure]
         The mind is composed of three structures:
         \begin{descriptionlist}

src/year1/cognition-and-neuroscience/module2/sections/_deep_rl.tex

@@ -3,7 +3,7 @@
 
 Deep learning has a large impact on neuroscience. For instance:
 \begin{casestudy}[Neural network to represent parietal neurons \cite{parietal_network}]
-    A monkey is tasked to maintain fixation at a point with a stimulus visible within its receptive field. The fixation point and the stimulus are moved together so that, for the retina, the stimulus is at the same retinal coordinate.
+    A monkey is tasked to maintain fixation at a point with a stimulus visible within its receptive field. The fixation point and the stimulus are moved together so that, for the retina, the stimulus is at the same retinal coordinates.
 
     \begin{figure}[H]
         \centering
@@ -116,7 +116,7 @@ There are at least two factors that cause the sample inefficiency of deep reinfo
         \end{remark}
 
     \item[Long-term memory] \marginnote{Long-term memory}
-        Inactive past memories that have to be reactivated to use them. In humans, it comes in different forms:
+        Inactive past memories that have to be reactivated in order to use them. In humans, it comes in different forms:
         \begin{descriptionlist}
             \item[Declarative/Explicit]
                 Memory that can be expressed in a propositional form (i.e. by words).
@@ -138,7 +138,7 @@ There are at least two factors that cause the sample inefficiency of deep reinfo
         \end{figure}
 
         \begin{remark}
-            On the other spectrum, there are short-term and working memories.
+            On the other spectrum, there is short-term and working memory.
             The former, once formed, is sort of read-only (e.g. repeat a sequence of numbers). The latter allows the manipulation of its content (e.g. repeat a sequence of numbers in reverse order).
         \end{remark}
 
@@ -212,7 +212,7 @@ There are at least two factors that cause the sample inefficiency of deep reinfo
 
         \begin{casestudy}[Replay in rats]
             The movements and the activity of a neuron (place cell) in the hippocampus of a rat are recorded.
-            It has been observed that the place cell fires only at a specific spatial area (and it changes if the rat is moved to another environment).
+            It has been observed that the place cell fires only at a specific spatial area (which changes if the rat is moved to another environment).
             \begin{figure}[H]
                 \centering
                 \includegraphics[width=0.55\linewidth]{./img/hippocampus_replay1.png}
@@ -220,14 +220,21 @@ There are at least two factors that cause the sample inefficiency of deep reinfo
 
             \indenttbox
             \begin{remark}
-                Grid cells are another type of cells in the medial entorhinal cortex that fire based on spatial features.
-                Differently from place cells, this type of cell is active in multiple zones arranged in a regular manner (i.e. triangular or hexagonal units).
+                \phantom{}\\[0.5em]
+                \begin{minipage}{0.6\linewidth}
+                    Grid cells are another type of cells in the medial entorhinal cortex that fire based on spatial features.
+                    Differently from place cells, this type of cell is active in multiple zones arranged in a regular manner (i.e. triangular or hexagonal units).
+    
+                    In other words, place cells are based on landmarks and grid cells are based on self-motion.
+                    
+                \end{minipage}
+                \begin{minipage}{0.3\linewidth}
+                    \begin{figure}[H]
+                        \centering
+                        \includegraphics[width=0.9\linewidth]{./img/grid_place_cell.png}
+                    \end{figure}
+                \end{minipage}
 
-                In other words, place cells are based on landmarks and grid cells are based on self-motion.
-                \begin{figure}[H]
-                    \centering
-                    \includegraphics[width=0.25\linewidth]{./img/grid_place_cell.png}
-                \end{figure}
             \end{remark}
 
             By placing the rat on a triangular track where some reward is available, a correlation between two place cells have been recorded.
@@ -247,7 +254,7 @@ There are at least two factors that cause the sample inefficiency of deep reinfo
             \phantom{}
             \begin{description}
                 \item[Experiment 1] 
-                    In each trial, candidates are asked to choose a slot machine to spin (bandit problem). Each slot has a different point reward that changes at each trial.
+                    In each trial, candidates are asked to choose a slot machine to spin (bandit problem). Each slot has a different reward in points that changes at each trial.
                     \begin{figure}[H]
                         \centering
                         \includegraphics[width=0.55\linewidth]{./img/memory_decision1.png}
@@ -294,7 +301,7 @@ There are at least two factors that cause the sample inefficiency of deep reinfo
     Monkeys are presented with two unfamiliar objects and are asked to grab one of them.
     Under an object lays either food or nothing.
     The same procedure is repeated for six trials and the position of the two objects can be swapped.
-    After the round of trials, new rounds are repeated with new objects.
+    After the first round of trials, new rounds are repeated with new objects.
 
     It has been observed that after some rounds, monkeys are able to learn the task in one shot.
 

src/year1/cognition-and-neuroscience/module2/sections/_dopamine.tex

@@ -360,7 +360,7 @@ Despite that, dopamine still seems to integrate predictive information from mode
 
 \begin{casestudy}[Monkey saccade \cite{saccade}]
     Monkeys are required to solve a memory-guided saccade task where, after fixation, 
-    a light is flashed in one of the four directions indicating the saccade to be made after the fixation point went off.
+    a light is flashed in one of the four directions indicating the saccade to be made after the fixation point goes off.
     \begin{figure}[H]
         \centering
         \includegraphics[width=0.4\linewidth]{./img/saccade1.png}
@@ -379,9 +379,9 @@ Despite that, dopamine still seems to integrate predictive information from mode
 
     It is expected that the animal's reward prediction increases after each non-rewarded trial.
     In other words, as the reward is more likely after each non-rewarded trial, positive prediction error should decrease and 
-    negative prediction error should be stronger (i.e. decrease).
+    negative prediction error should be stronger.
 
-    Results show that dopamine neurons are less active if the reward is delivered
+    Results show that dopamine neurons are less active if the reward is delivered later
     and more depressed if the reward is omitted after each non-rewarded trial.
     \begin{figure}[H]
         \centering
@@ -389,7 +389,7 @@ Despite that, dopamine still seems to integrate predictive information from mode
     \end{figure}
 
     The results are in contrast with an exclusive model-free view of dopamine as, if this were the case, 
-    learning would only involve past non-rewarded trials causing positive prediction error to decrease and negative prediction error to be weaker (i.e. increase).
+    learning would only involve past non-rewarded trials causing positive prediction error to decrease and negative prediction error to be weaker.
     Therefore, dopamine might process prediction error in both model-free and model-based approaches.
 \end{casestudy}
 

src/year1/cognition-and-neuroscience/module2/sections/_object_recognition.tex

@@ -79,7 +79,7 @@
 
                 \begin{casestudy}[Agnosia]
                     Patients with agnosia have their last level of vision damaged.
-                    They can see (e.g. avoid obstacles) but cannot recognize object or get easily confused.
+                    They can see (e.g. avoid obstacles) but cannot recognize objects or get easily confused.
                 \end{casestudy}
         \end{descriptionlist}
 
@@ -183,6 +183,7 @@
         \begin{remark}
             Neurons might only react to a specific type of stimuli in the receptive field (e.g. color, direction, \dots).
 
+            \indenttbox
             \begin{casestudy}
                 It has been seen that a neuron fires only if a stimulus is presented in its receptive field while moving upwards.
             \end{casestudy}
@@ -218,7 +219,7 @@
                 have a visual angle of about $0.1^\circ$ while the neurons at the visual periphery reach up to $1^\circ$ of visual angle.
 
                 Accordingly, more cortical space is dedicated to the central part of the visual field.
-                This densely packed amount of smaller receptive fields allows for the highest spatial resolution at the center of the visual field.
+                This densely packed amount of smaller receptive fields allows to obtain the highest spatial resolution at the center of the visual field.
 
                 \begin{figure}[H]
                     \centering
@@ -359,7 +360,7 @@ This is the result of the alignment of different circular receptive fields of th
 \marginnote{Complex cells}
 
 Neurons with a rectangular receptive field larger than simple cells.
-They respond to linear stimuli with a specific orientation and move in a particular direction (position invariance).
+They respond to linear stimuli with a specific orientation and with a specific movement direction (position invariance).
 
 \begin{remark}
     At this stage, the position of the stimulus is not relevant anymore as the ON and OFF zones of the previous cells are mixed.
@@ -369,7 +370,7 @@ They respond to linear stimuli with a specific orientation and move in a particu
     \centering
     \begin{subfigure}{0.45\linewidth}
         \centering
-        \includegraphics[width=0.9\linewidth]{./img/complex_cell.png}
+        \includegraphics[width=0.85\linewidth]{./img/complex_cell.png}
     \end{subfigure}
     \begin{subfigure}{0.45\linewidth}
         \centering

src/year1/image-processing-and-computer-vision/module2/sections/_image_formation.tex

@@ -711,7 +711,7 @@ Therefore, the complete workflow for image formation becomes the following:
         \end{remark}
 
     \item[Initial distortion parameters guess]
-        The current estimate of the homographies $\matr{H}_i$ project WRF points into ideal (undistorted) IRF coordinates $\vec{m}_\text{undist}$.
+        The current estimate of the homographies $\matr{H}_i$ project WRF points into ideal (undistorted) IRF coordinates $\vec{m}_\text{undist}$ (this is an assumption).
         On the other hand, the coordinates $\vec{m}$ of the corners in the actual image are distorted.
 
         The original algorithm estimates the parameters of the radial distortion function defined as: