diff --git a/01_overview-getting-familiar-with-torch.Rmd b/01_overview-getting-familiar-with-torch.Rmd
index d155f29..a19a892 100644
--- a/01_overview-getting-familiar-with-torch.Rmd
+++ b/01_overview-getting-familiar-with-torch.Rmd
@@ -4,6 +4,13 @@
 
 - THESE ARE NICE TO HAVE BUT NOT ABSOLUTELY NECESSARY
 
+## Setup {-}
+
+```{r install-torch}
+Sys.setenv(TORCH_INSTALL = 1)
+library(torch)
+```
+
 ## SLIDE 1 {-}
 
 - ADD SLIDES AS SECTIONS (`##`).
diff --git a/03_tensors.Rmd b/03_tensors.Rmd
index 3d16e0d..0cfcf64 100644
--- a/03_tensors.Rmd
+++ b/03_tensors.Rmd
@@ -2,12 +2,406 @@
 
 **Learning objectives:**
 
-- THESE ARE NICE TO HAVE BUT NOT ABSOLUTELY NECESSARY
+Learn how to:
+
+- Create and modify tensors
+- Access parts of tensors
+- Apply operations to tensors
+
+## What are tensors? {-}
+
+From the book:
+
+> "... *tensors* are 'just' multi-dimensional arrays optimized for fast computation - 
+not on the CPU only but also on specialized devices such as GPUs and TPUs.
+
+Load libraries:
+
+```{r message = FALSE, warning=FALSE}
+library(torch)
+library(dplyr)
+library(ggplot2)
+```
+
+Create a tensor:
+
+```{r}
+t1 <- torch_tensor(2)
+t1$shape
+```
+
+Parameters of `torch_tensor()` (see help file for this function):
+
+* `data`
+* `dtype`
+* `device`
+* `requires_grad`
+* `pin_memory`
+
+Look at the attributes of the tensor:
+
+```{r}
+t1$dtype
+t1$device
+t1$shape
+summary(t1)
+```
+
+Change attributes:
+
+```{r}
+# to integer:
+t2 <- t1$to(dtype = torch_int())
+t2$dtype
+```
+
+```{r eval=FALSE}
+# utilize GPU:
+t2 <- t1$to(device = "cuda")
+t2$device
+```
+
+
+## Creating tensors
+
+### Tensors from values
+
+**Defaults**: long integer type and CPU device
+
+Let's test this:
+
+```{r}
+torch_tensor(1:2)
+```
+In the example above, the type is long integer.
+
+```{r}
+torch_tensor(c(1,2))
+```
+
+In this example, the type is float instead of a long integer. 
+My guess is that the difference is due to the difference in data types of `1:2` and `c(1,2)`:
+
+```{r}
+class(c(1,2))
+class(1:2)
+```
+
+Explicitly set the type and device:
+
+```{r eval = FALSE}
+torch_tensor(1:5, dtype = torch_float(), device = "cuda")
+```
+
+Two-dimensional tensor:
+
+```{r}
+torch_tensor(matrix(1:9, ncol = 3, byrow = TRUE))
+```
+
+Higher dimensions:
+
+```{r}
+torch_tensor(array(1:24, dim = c(4,3,2)))
+```
+
+### Tensors from specifications
+
+```{r}
+#normal distribution:
+torch_randn(3, 3)
+
+#unfiorm distribution
+torch_rand(3, 3)
+```
+
+```{r}
+torch_zeros(2,2)
+torch_ones(3,3)
+torch_eye(3,3)
+torch_diag(c(1,2,3))
+```
+
+See full list at https://torch.mlverse.org/docs/reference/#tensor-creation-utilities. 
+
+### Tensors from datasets
+
+We'll look at the `presidential` dataset from the `ggplot2` package:
+
+```{r}
+data(presidential) # from ggplot2 
+presidential |> glimpse()
+```
+
+```{r}
+presidential |> 
+  mutate(name = as.numeric(factor(name)),
+         start = as.numeric(start),
+         end = as.numeric(end),
+         party = as.numeric(factor(party))) |>
+  as.matrix() |>
+  torch_tensor() |>
+  print(5)
+```
+
+If your data contains `NA`'s, you'll need to convert them before training a neural network 
+(topic to be covered later).
+
+## Operations on Tensors
+
+Define two tensors:
+
+```{r}
+t1 <- torch_tensor(c(1,2))
+t2 <- torch_tensor(c(3,4))
+```
+
+Add:
+
+```{r}
+torch_add(t1, t2) # t1 is NOT modified
+t1$add(t2) # t1 is NOT modified
+t1$add_(t2) # t1 IS modified
+t1
+```
+
+In general: 
+
+* underscore appended to operation indicates modification in-place.
+* torch does not distinguish between row and column vectors.
+* Other examples of operations: `torch_t()`, `torch_dot()`, `torch_matmul()`, and `torch_multiply()`. 
+See https://torch.mlverse.org/docs/reference/#mathematical-operations-on-tensors.
+
+Another example:
+
+```{r}
+t1 <- torch_tensor(1:3)
+t2 <- torch_tensor(4:6)
+```
+
+```{r}
+t1$dot(t2)
+```
+
+### Summary operations
+
+Create a matrix and a tensor using outer products:
+
+```{r}
+m <- outer(1:3, 1:6)
+t <- torch_outer(torch_tensor(1:3), torch_tensor(1:6))
+```
+
+```{r}
+apply(m, 1, sum) # row sums (of R matrix)
+t$sum(dim = 2) # row sums (of tensor)
+```
+
+* In R, we *group* by row (dimension 1) for row summaries and by columns (dimension 2) for column summaries
+* In `torch, we *collapse* the columns (dimension 2) for row summaries and the rows (dimension 1) for column summaries.
+
+Time series example: Two features collected three times for four individuals. 
+
+* Dimension 1: Runs over individuals
+* Dimension 2: Runs over points in time
+* Dimension 3: Runs over features
+
+```{r}
+t <- torch_randn(4, 3, 2)
+t
+```
+
+Obtain averages of features, independent of subject (dimension 1) and time (dimension 2):
+
+```{r}
+t$mean(dim = c(1, 2))
+```
+
+
+## Accessing Parts of a Tensor (indexing and slicing)
+
+Indexing in `torch` is 1-based. Very similar to functionality in R.
+
+Example:
+
+```{r}
+t <- torch_tensor(matrix(1:9, ncol = 3, byrow = TRUE))
+t
+t[1, ] # row 1. 
+t[1, , drop = FALSE] # Same as above except that dimensionality is preserved
+```
+
+Slicing example:
+
+```{r}
+t <- torch_rand(3, 3, 3)
+t[1:2, 2:3, c(1, 3)]
+```
+
+### Beyond R
+
+Access last element:
+
+```{r}
+t <- torch_tensor(matrix(1:4, ncol = 2, byrow = TRUE))
+t[-1, -1]
+```
+
+```{r}
+t <- torch_tensor(1:4)
+t[-1]
+```
+
+Compare to R:
+
+```{r}
+# matrix:
+m <- matrix(1:4, ncol = 2, byrow = TRUE)
+m[-1, -1]
+
+# vector:
+m <- 1:4
+m[-1]
+```
+
+Step pattern:
+
+```{r}
+t <- torch_tensor(matrix(1:20, ncol = 10, byrow = TRUE))
+t
+t[ , 1:8:2] # every other value in columns 1 through 8
+```
+
+Use `..` to designate all dimensions not explicitly referenced.
+
+```{r}
+t2 <- torch_randn(2, 2, 2)
+t2
+t2[2, ..] # 2nd element of the 1st dimension
+```
+
+## Reshaping Tensors
+
+Two methods:
+
+* `view`: Zero-copy reshape (by changing tensor metadata). Will fail if zero-copy reshape is not possible.
+* `reshape`: Uses zero-copy reshape (via metadata) when possible. If not, then will make a copy.
+
+```{r}
+t <- torch_tensor(matrix(1:15, nrow = 3, byrow = TRUE))
+t
+t$stride()
+```
+
+`$stride()` tells us the jump necessary to go from one element to the next in a single dimension (e.g. the stride necessary to get from one row to the next).
+
+Next, change shape of `t` using `view()`:
+
+```{r}
+t2 <- t$view(c(5,3))
+t2
+t2$stride()
+```
+
+
+Check memory location:
+
+```{r}
+t$storage()$data_ptr()
+t2$storage()$data_ptr() # same location
+```
+
+
+`view()` vs `reshape()`:
+
+When two operations that change the stride are done in sequence, this will likely fail.
+Below is an example of transpose, `t()` followed by `view()`.
+
+```{r eval = FALSE}
+t$t()$view(15) # error
+```
+
+However, `reshape()` makes a copy of the underlying data and does not fail.
+
+```{r}
+t3 <- t$t()$reshape(15) # no error
+t3
+```
+
+
+Now the memory locations of t3 and the original data are different:
+
+```{r}
+t$storage()$data_ptr()
+t3$storage()$data_ptr() # different location
+```
+
+Two additional functions that are zero-copy operations:
+
+* `squeeze()`: Removes singleton dimensions
+* `unsqueeze()`: Adds a singleton dimension
+
+```{r}
+t <- torch_rand(3) # one dimensional tensor of shape 3
+t
+t2 <- t$unsqueeze(1) # two dimensional tensor of shape {1,3}
+t2
+```
+Inspect location:
+
+```{r}
+t$storage()$data_ptr()
+t2$storage()$data_ptr() # same location
+```
+
+
+## Broadcasting
+
+Example: Want to add two tensors of shape 3x7x1 and 1x5
+
+```
+t1 shape: 3 7 1
+t2 shape:   1 5
+```
+
+Broadcast (from right to left):
+
+```
+t1 shape: 3 7 5
+t2 shape:   7 5
+```
+
+Next, virtual expansion:
+
+```
+t1 shape: 3 7 5
+t2 shape: 1 7 5
+```
+
+And, broadcast again:
+
+```
+t1 shape: 3 7 5
+t2 shape: 3 7 5
+```
+
+Let's see it in action:
+
+```{r}
+t1 <- torch_ones(3, 7, 1)
+t2 <- torch_zeros(1, 5)
+torch_add(t1, t2)
+```
+
+The following will NOT work:
+
+```{r eval = FALSE}
+t1 <- torch_ones(3, 7, 1)
+t2 <- torch_zeros(6, 5)
+torch_add(t1, t2)
+```
 
-## SLIDE 1 {-}
 
-- ADD SLIDES AS SECTIONS (`##`).
-- TRY TO KEEP THEM RELATIVELY SLIDE-LIKE; THESE ARE NOTES, NOT THE BOOK ITSELF.
 
 ## Meeting Videos {-}
 
diff --git a/DESCRIPTION b/DESCRIPTION
index 7dd5b88..f9dc574 100644
--- a/DESCRIPTION
+++ b/DESCRIPTION
@@ -9,5 +9,8 @@ Depends:
     R (>= 3.1.0)
 Imports: 
     bookdown,
-    rmarkdown
+    dplyr,
+    ggplot2,
+    rmarkdown,
+    torch
 Encoding: UTF-8