Merge pull request #386 from neuromatch/W2D1-hotfix

W2D1

tutorials/W2D1_Macrocircuits/W2D1_Tutorial3.ipynb

@@ -53,7 +53,7 @@
     "4. Use neural decoding to understand why the modular architecture for this specific task is advantageous.\n",
     "5. Keep the No-Free-Lunch Theorem in mind: the benefits of a modular architecture for one task cannot apply to all possible tasks.\n",
     "\n",
-    "This tutorial is based on this [paper](https://www.biorxiv.org/content/10.1101/2022.12.07.519515v2.abstract)."
+    "This tutorial is based on this [paper](https://www.science.org/doi/10.1126/sciadv.adk1256)."
    ]
   },
   {
@@ -3340,7 +3340,7 @@
     "---\n",
     "# Bonus Section 2: Generalization, but no free lunch\n",
     "\n",
-    "The No Free Lunch theorems proved that no inductive bias can excel across all tasks. It has been studied in the [paper](https://www.biorxiv.org/content/10.1101/2022.12.07.519515v2.abstract) that agents with a modular architecture can acquire the underlying structure of the training task. In contrast, holistic agents tend to acquire different knowledge than modular agents during training, such as forming beliefs based on unreliable information sources or exhibiting less efficient control actions. The novel gain task has a structure similar to the training task, consequently, a modular agent that accurately learns the training task's structure can leverage its knowledge in these novel tasks.\n",
+    "The No Free Lunch theorems proved that no inductive bias can excel across all tasks. It has been studied in the [paper](https://www.science.org/doi/10.1126/sciadv.adk1256) that agents with a modular architecture can acquire the underlying structure of the training task. In contrast, holistic agents tend to acquire different knowledge than modular agents during training, such as forming beliefs based on unreliable information sources or exhibiting less efficient control actions. The novel gain task has a structure similar to the training task, consequently, a modular agent that accurately learns the training task's structure can leverage its knowledge in these novel tasks.\n",
     "\n",
     "However, it is worth noting that an infinite number of new tasks can be constructed, diverging from the training task's structure but aligning with the 'inferior' beliefs and control acquired by holistic agents.\n",
     "\n",

tutorials/W2D1_Macrocircuits/further_reading.md

@@ -13,4 +13,4 @@
 
 ## Tutorial 3: Modularity
 
-- [Inductive biases of neural networks for generalization in spatial navigation](https://www.biorxiv.org/content/10.1101/2022.12.07.519515v1)
+- [Inductive biases of neural network modularity in spatial navigation](https://www.science.org/doi/10.1126/sciadv.adk1256)

tutorials/W2D1_Macrocircuits/instructor/W2D1_Tutorial3.ipynb

@@ -53,7 +53,7 @@
     "4. Use neural decoding to understand why the modular architecture for this specific task is advantageous.\n",
     "5. Keep the No-Free-Lunch Theorem in mind: the benefits of a modular architecture for one task cannot apply to all possible tasks.\n",
     "\n",
-    "This tutorial is based on this [paper](https://www.biorxiv.org/content/10.1101/2022.12.07.519515v2.abstract)."
+    "This tutorial is based on this [paper](https://www.science.org/doi/10.1126/sciadv.adk1256)."
    ]
   },
   {
@@ -3350,7 +3350,7 @@
     "---\n",
     "# Bonus Section 2: Generalization, but no free lunch\n",
     "\n",
-    "The No Free Lunch theorems proved that no inductive bias can excel across all tasks. It has been studied in the [paper](https://www.biorxiv.org/content/10.1101/2022.12.07.519515v2.abstract) that agents with a modular architecture can acquire the underlying structure of the training task. In contrast, holistic agents tend to acquire different knowledge than modular agents during training, such as forming beliefs based on unreliable information sources or exhibiting less efficient control actions. The novel gain task has a structure similar to the training task, consequently, a modular agent that accurately learns the training task's structure can leverage its knowledge in these novel tasks.\n",
+    "The No Free Lunch theorems proved that no inductive bias can excel across all tasks. It has been studied in the [paper](https://www.science.org/doi/10.1126/sciadv.adk1256) that agents with a modular architecture can acquire the underlying structure of the training task. In contrast, holistic agents tend to acquire different knowledge than modular agents during training, such as forming beliefs based on unreliable information sources or exhibiting less efficient control actions. The novel gain task has a structure similar to the training task, consequently, a modular agent that accurately learns the training task's structure can leverage its knowledge in these novel tasks.\n",
     "\n",
     "However, it is worth noting that an infinite number of new tasks can be constructed, diverging from the training task's structure but aligning with the 'inferior' beliefs and control acquired by holistic agents.\n",
     "\n",

tutorials/W2D1_Macrocircuits/student/W2D1_Tutorial3.ipynb

@@ -53,7 +53,7 @@
     "4. Use neural decoding to understand why the modular architecture for this specific task is advantageous.\n",
     "5. Keep the No-Free-Lunch Theorem in mind: the benefits of a modular architecture for one task cannot apply to all possible tasks.\n",
     "\n",
-    "This tutorial is based on this [paper](https://www.biorxiv.org/content/10.1101/2022.12.07.519515v2.abstract)."
+    "This tutorial is based on this [paper](https://www.science.org/doi/10.1126/sciadv.adk1256)."
    ]
   },
   {
@@ -3000,7 +3000,7 @@
     "---\n",
     "# Bonus Section 2: Generalization, but no free lunch\n",
     "\n",
-    "The No Free Lunch theorems proved that no inductive bias can excel across all tasks. It has been studied in the [paper](https://www.biorxiv.org/content/10.1101/2022.12.07.519515v2.abstract) that agents with a modular architecture can acquire the underlying structure of the training task. In contrast, holistic agents tend to acquire different knowledge than modular agents during training, such as forming beliefs based on unreliable information sources or exhibiting less efficient control actions. The novel gain task has a structure similar to the training task, consequently, a modular agent that accurately learns the training task's structure can leverage its knowledge in these novel tasks.\n",
+    "The No Free Lunch theorems proved that no inductive bias can excel across all tasks. It has been studied in the [paper](https://www.science.org/doi/10.1126/sciadv.adk1256) that agents with a modular architecture can acquire the underlying structure of the training task. In contrast, holistic agents tend to acquire different knowledge than modular agents during training, such as forming beliefs based on unreliable information sources or exhibiting less efficient control actions. The novel gain task has a structure similar to the training task, consequently, a modular agent that accurately learns the training task's structure can leverage its knowledge in these novel tasks.\n",
     "\n",
     "However, it is worth noting that an infinite number of new tasks can be constructed, diverging from the training task's structure but aligning with the 'inferior' beliefs and control acquired by holistic agents.\n",
     "\n",