Prepod day 1

tutorials/W1D1_Generalization/W1D1_Tutorial2.ipynb

@@ -910,7 +910,17 @@
     "* The inputs are 16-dimensional, with 15 dimensions corresponding to the reach condition and the 16th dimension corresponding to the go-cue.\n",
     "* The outputs are 2-dimensional, corresponding to the target electromyography (EMG) data for **2** muscles.\n",
     "\n",
-    "The inputs are 16-dimensional and indicate to the model **which reach to perform**. Each reach condition is coded as a random 15-dimensional vector which is held constant during the movement preparation period from -1s to 0s. The go-cue in the 16th dimension has a small bump at 0s, indicating to the model that it should generate the output. \n",
+    "The inputs to the model are 16-dimensional vectors that specify which reach to perform. Each reach condition is represented by a unique 15-dimensional vector, which remains constant during the movement preparation period from -1s to 0s. The 16th dimension, a go-cue, features a small bump at 0s, signaling the model to generate the output.\n",
+    "\n",
+    "Let's clarify the nature of these inputs:\n",
+    "\n",
+    "These inputs inform the Recurrent Neural Network (RNN) about the type of reach (condition) being executed. Each condition may involve reaching in different directions or towards various targets. These conditions are crucial because they enable the RNN to distinguish between different types of reaches, thus allowing it to predict muscle activity accurately for each specific condition.\n",
+    "\n",
+    "Additionally, there is a condition-independent hold cue among the 15 inputs, consistent across all conditions. This cue indicates a period during which the model should maintain its state before initiating the reach.\n",
+    "\n",
+    "To illustrate, consider a robot arm that needs to reach different objects positioned in various locations. The condition-specific inputs would guide the robot on which object to reach for (e.g., left, right, up, down). Prior to each reach, the robot's \"brain\" (neural activity) prepares in a particular manner, and this readiness is recorded. These readiness signals are then fed into the RNN, helping it understand and prepare for the specific reach.\n",
+    "\n",
+    "Initially, the robot remains still (hold cue). Following a brief delay, it receives a signal to begin the reach. These delays and specific inputs train the robot to accurately perform the reach for each condition.\n",
     "\n",
     "Let's look at the few of the inputs and outputs to get a better understanding of the data."
    ]

tutorials/W1D1_Generalization/instructor/W1D1_Tutorial2.ipynb

@@ -910,7 +910,17 @@
     "* The inputs are 16-dimensional, with 15 dimensions corresponding to the reach condition and the 16th dimension corresponding to the go-cue.\n",
     "* The outputs are 2-dimensional, corresponding to the target electromyography (EMG) data for **2** muscles.\n",
     "\n",
-    "The inputs are 16-dimensional and indicate to the model **which reach to perform**. Each reach condition is coded as a random 15-dimensional vector which is held constant during the movement preparation period from -1s to 0s. The go-cue in the 16th dimension has a small bump at 0s, indicating to the model that it should generate the output. \n",
+    "The inputs to the model are 16-dimensional vectors that specify which reach to perform. Each reach condition is represented by a unique 15-dimensional vector, which remains constant during the movement preparation period from -1s to 0s. The 16th dimension, a go-cue, features a small bump at 0s, signaling the model to generate the output.\n",
+    "\n",
+    "Let's clarify the nature of these inputs:\n",
+    "\n",
+    "These inputs inform the Recurrent Neural Network (RNN) about the type of reach (condition) being executed. Each condition may involve reaching in different directions or towards various targets. These conditions are crucial because they enable the RNN to distinguish between different types of reaches, thus allowing it to predict muscle activity accurately for each specific condition.\n",
+    "\n",
+    "Additionally, there is a condition-independent hold cue among the 15 inputs, consistent across all conditions. This cue indicates a period during which the model should maintain its state before initiating the reach.\n",
+    "\n",
+    "To illustrate, consider a robot arm that needs to reach different objects positioned in various locations. The condition-specific inputs would guide the robot on which object to reach for (e.g., left, right, up, down). Prior to each reach, the robot's \"brain\" (neural activity) prepares in a particular manner, and this readiness is recorded. These readiness signals are then fed into the RNN, helping it understand and prepare for the specific reach.\n",
+    "\n",
+    "Initially, the robot remains still (hold cue). Following a brief delay, it receives a signal to begin the reach. These delays and specific inputs train the robot to accurately perform the reach for each condition.\n",
     "\n",
     "Let's look at the few of the inputs and outputs to get a better understanding of the data."
    ]

tutorials/W1D1_Generalization/student/W1D1_Tutorial2.ipynb

@@ -910,7 +910,17 @@
     "* The inputs are 16-dimensional, with 15 dimensions corresponding to the reach condition and the 16th dimension corresponding to the go-cue.\n",
     "* The outputs are 2-dimensional, corresponding to the target electromyography (EMG) data for **2** muscles.\n",
     "\n",
-    "The inputs are 16-dimensional and indicate to the model **which reach to perform**. Each reach condition is coded as a random 15-dimensional vector which is held constant during the movement preparation period from -1s to 0s. The go-cue in the 16th dimension has a small bump at 0s, indicating to the model that it should generate the output. \n",
+    "The inputs to the model are 16-dimensional vectors that specify which reach to perform. Each reach condition is represented by a unique 15-dimensional vector, which remains constant during the movement preparation period from -1s to 0s. The 16th dimension, a go-cue, features a small bump at 0s, signaling the model to generate the output.\n",
+    "\n",
+    "Let's clarify the nature of these inputs:\n",
+    "\n",
+    "These inputs inform the Recurrent Neural Network (RNN) about the type of reach (condition) being executed. Each condition may involve reaching in different directions or towards various targets. These conditions are crucial because they enable the RNN to distinguish between different types of reaches, thus allowing it to predict muscle activity accurately for each specific condition.\n",
+    "\n",
+    "Additionally, there is a condition-independent hold cue among the 15 inputs, consistent across all conditions. This cue indicates a period during which the model should maintain its state before initiating the reach.\n",
+    "\n",
+    "To illustrate, consider a robot arm that needs to reach different objects positioned in various locations. The condition-specific inputs would guide the robot on which object to reach for (e.g., left, right, up, down). Prior to each reach, the robot's \"brain\" (neural activity) prepares in a particular manner, and this readiness is recorded. These readiness signals are then fed into the RNN, helping it understand and prepare for the specific reach.\n",
+    "\n",
+    "Initially, the robot remains still (hold cue). Following a brief delay, it receives a signal to begin the reach. These delays and specific inputs train the robot to accurately perform the reach for each condition.\n",
     "\n",
     "Let's look at the few of the inputs and outputs to get a better understanding of the data."
    ]