Process tutorial notebooks

neuromatch · Aug 7, 2024 · bfa1e48 · bfa1e48
1 parent ecc5837
commit bfa1e48
Show file tree

Hide file tree

Showing 6 changed files with 103 additions and 50 deletions.
diff --git a/tutorials/W1D2_ComparingTasks/W1D2_Tutorial2.ipynb b/tutorials/W1D2_ComparingTasks/W1D2_Tutorial2.ipynb
@@ -57,6 +57,7 @@
    "id": "aceddeef-ea13-4224-8ca8-70563edbcf00",
    "metadata": {
     "cellView": "form",
+    "execution": {},
     "tags": [
      "remove-input"
     ]
@@ -106,6 +107,7 @@
    "id": "f3eeb07c-5de3-428e-a422-3a58f1124b43",
    "metadata": {
     "cellView": "form",
+    "execution": {},
     "tags": [
      "hide-input"
     ]
@@ -137,6 +139,7 @@
    "id": "b56685b9-3f17-4a55-acca-73dad5623992",
    "metadata": {
     "cellView": "form",
+    "execution": {},
     "tags": [
      "hide-input"
     ]
@@ -179,6 +182,7 @@
    "id": "cebc5798-a8dd-4985-b0ec-d5facf4ee700",
    "metadata": {
     "cellView": "form",
+    "execution": {},
     "tags": [
      "hide-input"
     ]
@@ -200,6 +204,7 @@
    "id": "386b49a0-9409-496a-8656-579ca3e4af5f",
    "metadata": {
     "cellView": "form",
+    "execution": {},
     "tags": [
      "hide-input"
     ]
@@ -270,7 +275,8 @@
    "execution_count": null,
    "id": "a47d3240-0a70-4e3b-af13-2c02528867fa",
    "metadata": {
-    "cellView": "form"
+    "cellView": "form",
+    "execution": {}
    },
    "outputs": [],
    "source": [
@@ -327,7 +333,8 @@
    "execution_count": null,
    "id": "2cb688bf-38c8-40e1-b40b-c4014cdaffba",
    "metadata": {
-    "cellView": "form"
+    "cellView": "form",
+    "execution": {}
    },
    "outputs": [],
    "source": [
@@ -378,7 +385,9 @@
    "cell_type": "code",
    "execution_count": null,
    "id": "d7c432bf-1de5-43f6-b887-087d17e9ec0f",
-   "metadata": {},
+   "metadata": {
+    "execution": {}
+   },
    "outputs": [],
    "source": [
     "class ResidualBlock(nn.Module):\n",
@@ -428,7 +437,9 @@
    "cell_type": "code",
    "execution_count": null,
    "id": "8cdc7833-aafd-4ee0-83fd-af9922dd5857",
-   "metadata": {},
+   "metadata": {
+    "execution": {}
+   },
    "outputs": [],
    "source": [
     "class Model(nn.Module):\n",
@@ -465,7 +476,9 @@
    "cell_type": "code",
    "execution_count": null,
    "id": "d72d9231-35a5-405e-b2f6-135e134e0ae9",
-   "metadata": {},
+   "metadata": {
+    "execution": {}
+   },
    "outputs": [],
    "source": [
     "# Define the transformations for the MNIST dataset\n",
@@ -509,7 +522,9 @@
    "cell_type": "code",
    "execution_count": null,
    "id": "ffd3ab89-ffd9-408c-8396-25b88e0930fb",
-   "metadata": {},
+   "metadata": {
+    "execution": {}
+   },
    "outputs": [],
    "source": [
     "# Initialize the model with specified input, output, and hidden dimensions\n",
@@ -541,7 +556,9 @@
    "cell_type": "code",
    "execution_count": null,
    "id": "b2f9122b-ce83-464f-b6c9-5af2feceac62",
-   "metadata": {},
+   "metadata": {
+    "execution": {}
+   },
    "outputs": [],
    "source": [
     "# First try with untrained network, find the cosine similarities within a class and across classes\n",
@@ -600,7 +617,9 @@
    "cell_type": "code",
    "execution_count": null,
    "id": "0JcR1VwliL1f",
-   "metadata": {},
+   "metadata": {
+    "execution": {}
+   },
    "outputs": [],
    "source": [
     "# Dictionary to store normalized embeddings for each class\n",
@@ -643,7 +662,9 @@
    "cell_type": "code",
    "execution_count": null,
    "id": "9594e7b0-e401-4792-8798-7f23109bcf94",
-   "metadata": {},
+   "metadata": {
+    "execution": {}
+   },
    "outputs": [],
    "source": [
     "# to_remove solution\n",
@@ -684,7 +705,8 @@
    "execution_count": null,
    "id": "26dd2a45-38ed-45ea-ad27-dfddfeb1613f",
    "metadata": {
-    "cellView": "form"
+    "cellView": "form",
+    "execution": {}
    },
    "outputs": [],
    "source": [
@@ -708,7 +730,9 @@
    "cell_type": "code",
    "execution_count": null,
    "id": "728682ab",
-   "metadata": {},
+   "metadata": {
+    "execution": {}
+   },
    "outputs": [],
    "source": [
     "# to_remove explanation\n",
@@ -729,7 +753,8 @@
    "execution_count": null,
    "id": "d13d5761-e2ec-4ae1-adb4-edbe57b77ba9",
    "metadata": {
-    "cellView": "form"
+    "cellView": "form",
+    "execution": {}
    },
    "outputs": [],
    "source": [
@@ -805,7 +830,9 @@
    "cell_type": "code",
    "execution_count": null,
    "id": "1bfe95da-cd87-4fea-bb22-c1419450f9cb",
-   "metadata": {},
+   "metadata": {
+    "execution": {}
+   },
    "outputs": [],
    "source": [
     "def dcl_loss(pos_pairs, neg_pairs, indices_tuple, temperature=0.07):\n",
@@ -970,7 +997,9 @@
    "cell_type": "code",
    "execution_count": null,
    "id": "67db2b80",
-   "metadata": {},
+   "metadata": {
+    "execution": {}
+   },
    "outputs": [],
    "source": [
     "# to_remove solution\n",
@@ -1140,7 +1169,9 @@
    "cell_type": "code",
    "execution_count": null,
    "id": "4d4f11e3-07e8-4104-9ae4-6a9fd6f037eb",
-   "metadata": {},
+   "metadata": {
+    "execution": {}
+   },
    "outputs": [],
    "source": [
     "def get_embeddings_labels(loader, model):\n",
@@ -1195,7 +1226,8 @@
    "execution_count": null,
    "id": "de3b7b2b-4351-4df8-8877-8f81b1d3878b",
    "metadata": {
-    "cellView": "form"
+    "cellView": "form",
+    "execution": {}
    },
    "outputs": [],
    "source": [
@@ -1219,7 +1251,9 @@
    "cell_type": "code",
    "execution_count": null,
    "id": "e88a8b40-faa7-4071-b070-0ca192e06d55",
-   "metadata": {},
+   "metadata": {
+    "execution": {}
+   },
    "outputs": [],
    "source": [
     "# Create DataLoader for the test dataset with a batch size of 50\n",
@@ -1332,7 +1366,9 @@
    "cell_type": "code",
    "execution_count": null,
    "id": "767e0f93-3f68-4db0-9edf-4891a3858f63",
-   "metadata": {},
+   "metadata": {
+    "execution": {}
+   },
    "outputs": [],
    "source": [
     "# Convert list of embeddings to a numpy array\n",
@@ -1365,7 +1401,9 @@
    "cell_type": "code",
    "execution_count": null,
    "id": "d4cdd26a-15c9-4c5f-82f7-4a65ca0bd5f7",
-   "metadata": {},
+   "metadata": {
+    "execution": {}
+   },
    "outputs": [],
    "source": [
     "# Convert list of embeddings to a numpy array\n",
@@ -1398,7 +1436,9 @@
    "cell_type": "code",
    "execution_count": null,
    "id": "2afe3c0d-7eed-4b2b-bdaa-bb258fc92d10",
-   "metadata": {},
+   "metadata": {
+    "execution": {}
+   },
    "outputs": [],
    "source": [
     "# Use t-SNE embeddings for visualization\n",
@@ -1454,7 +1494,9 @@
    "cell_type": "code",
    "execution_count": null,
    "id": "013D0NbXowjj",
-   "metadata": {},
+   "metadata": {
+    "execution": {}
+   },
    "outputs": [],
    "source": [
     "# Calculate the cosine similarity matrix between all the test images and the train images\n",
@@ -1492,7 +1534,9 @@
    "cell_type": "code",
    "execution_count": null,
    "id": "1ae6f966",
-   "metadata": {},
+   "metadata": {
+    "execution": {}
+   },
    "outputs": [],
    "source": [
     "best_idxs = np.argmax(sims_all, axis=1)\n",
@@ -1551,20 +1595,22 @@
    "cell_type": "code",
    "execution_count": null,
    "id": "fa73993e-bbaa-4d32-a7c1-3903c784883f",
-   "metadata": {},
+   "metadata": {
+    "execution": {}
+   },
    "outputs": [],
    "source": [
     "# to_remove explanation\n",
     "\"\"\"\n",
-    "Our brain's ability to perform contrastive learning is linked to the function of the ventral visual stream. \n",
-    "This area of the brain processes visual information and has been shown to develop hierarchical features that \n",
-    "capture the structure of visual input through self-supervised learning mechanisms. Evidence suggests that anterior \n",
-    "regions of the ventral visual stream, particularly the anterior occipito-temporal cortex (aOTC), encode substantial \n",
-    "information about object categories without requiring explicit category-level supervision (Konkle and Alvarez, 2022). \n",
-    "Instead, these representations emerge through domain-general learning from natural image structures, where the brain \n",
-    "differentiates between individual views and categories based on the inherent statistical properties of visual input \n",
-    "(Livingstone et al., 2019; Arcaro and Livingstone, 2021). This capability supports the notion that the brain's visual \n",
-    "system can form complex object representations and categorization using self-supervised learning frameworks similar to \n",
+    "Our brain's ability to perform contrastive learning is linked to the function of the ventral visual stream.\n",
+    "This area of the brain processes visual information and has been shown to develop hierarchical features that\n",
+    "capture the structure of visual input through self-supervised learning mechanisms. Evidence suggests that anterior\n",
+    "regions of the ventral visual stream, particularly the anterior occipito-temporal cortex (aOTC), encode substantial\n",
+    "information about object categories without requiring explicit category-level supervision (Konkle and Alvarez, 2022).\n",
+    "Instead, these representations emerge through domain-general learning from natural image structures, where the brain\n",
+    "differentiates between individual views and categories based on the inherent statistical properties of visual input\n",
+    "(Livingstone et al., 2019; Arcaro and Livingstone, 2021). This capability supports the notion that the brain's visual\n",
+    "system can form complex object representations and categorization using self-supervised learning frameworks similar to\n",
     "those in artificial neural networks.\n",
     "\"\"\""
    ]
@@ -1574,7 +1620,8 @@
    "execution_count": null,
    "id": "af3f6a4f-2185-41db-a451-09ade489790a",
    "metadata": {
-    "cellView": "form"
+    "cellView": "form",
+    "execution": {}
    },
    "outputs": [],
    "source": [

diff --git a/tutorials/W1D2_ComparingTasks/instructor/W1D2_Tutorial2.ipynb b/tutorials/W1D2_ComparingTasks/instructor/W1D2_Tutorial2.ipynb
@@ -1606,13 +1606,16 @@
    "source": [
     "# to_remove explanation\n",
     "\"\"\"\n",
-    "Konkle and Alvarez (2022) speculate that perceptual judgements of visual similarity might be\n",
-    "driven by a process of contrastive learning not unlike that in AI. Different views of the same\n",
-    "scene, like the augmentations used in contrastive learning, can be generated by eye movements,\n",
-    "body movements and the passage of time. The hippocampus could hold these views in working memory\n",
-    "to implement a form of contrastive learning in neocortex. These ideas are closely related to\n",
-    "long-standing ideas about predictive coding in the brain. This remains an active and fascinating\n",
-    "area of research.\n",
+    "Our brain's ability to perform contrastive learning is linked to the function of the ventral visual stream.\n",
+    "This area of the brain processes visual information and has been shown to develop hierarchical features that\n",
+    "capture the structure of visual input through self-supervised learning mechanisms. Evidence suggests that anterior\n",
+    "regions of the ventral visual stream, particularly the anterior occipito-temporal cortex (aOTC), encode substantial\n",
+    "information about object categories without requiring explicit category-level supervision (Konkle and Alvarez, 2022).\n",
+    "Instead, these representations emerge through domain-general learning from natural image structures, where the brain\n",
+    "differentiates between individual views and categories based on the inherent statistical properties of visual input\n",
+    "(Livingstone et al., 2019; Arcaro and Livingstone, 2021). This capability supports the notion that the brain's visual\n",
+    "system can form complex object representations and categorization using self-supervised learning frameworks similar to\n",
+    "those in artificial neural networks.\n",
     "\"\"\""
    ]
   },

diff --git a/tutorials/W1D2_ComparingTasks/solutions/W1D2_Tutorial2_Solution_0cd5a0dc.py b/tutorials/W1D2_ComparingTasks/solutions/W1D2_Tutorial2_Solution_0cd5a0dc.py
@@ -0,0 +1,12 @@
+"""
+Our brain's ability to perform contrastive learning is linked to the function of the ventral visual stream.
+This area of the brain processes visual information and has been shown to develop hierarchical features that
+capture the structure of visual input through self-supervised learning mechanisms. Evidence suggests that anterior
+regions of the ventral visual stream, particularly the anterior occipito-temporal cortex (aOTC), encode substantial
+information about object categories without requiring explicit category-level supervision (Konkle and Alvarez, 2022).
+Instead, these representations emerge through domain-general learning from natural image structures, where the brain
+differentiates between individual views and categories based on the inherent statistical properties of visual input
+(Livingstone et al., 2019; Arcaro and Livingstone, 2021). This capability supports the notion that the brain's visual
+system can form complex object representations and categorization using self-supervised learning frameworks similar to
+those in artificial neural networks.
+"""
diff --git a/tutorials/W1D2_ComparingTasks/solutions/W1D2_Tutorial2_Solution_81063c05.py b/tutorials/W1D2_ComparingTasks/solutions/W1D2_Tutorial2_Solution_81063c05.py
diff --git a/tutorials/W1D2_ComparingTasks/static/W1D2_Tutorial2_Solution_39154423_0.png b/tutorials/W1D2_ComparingTasks/static/W1D2_Tutorial2_Solution_39154423_0.png
diff --git a/tutorials/W1D2_ComparingTasks/student/W1D2_Tutorial2.ipynb b/tutorials/W1D2_ComparingTasks/student/W1D2_Tutorial2.ipynb
@@ -1413,7 +1413,7 @@
     "execution": {}
    },
    "source": [
-    "[*Click for solution*](https://github.com/neuromatch/NeuroAI_Course/tree/main/tutorials/W1D2_ComparingTasks/solutions/W1D2_Tutorial2_Solution_81063c05.py)\n",
+    "[*Click for solution*](https://github.com/neuromatch/NeuroAI_Course/tree/main/tutorials/W1D2_ComparingTasks/solutions/W1D2_Tutorial2_Solution_0cd5a0dc.py)\n",
     "\n"
    ]
   },