diff --git a/src/KSVD.jl b/src/KSVD.jl
index 898d011..f2d1f2a 100644
--- a/src/KSVD.jl
+++ b/src/KSVD.jl
@@ -87,7 +87,7 @@ To enable timing outputs, run `TimerOutputs.enable_debug_timings(KSVD)`.
 function ksvd(Y::AbstractMatrix{T}, n_atoms::Int, max_nnz=n_atoms÷10;
               ksvd_update_method = BatchedParallelKSVD{false, T}(; shuffle_indices=true, batch_size_per_thread=1),
               sparse_coding_method = ParallelMatchingPursuit(; max_nnz, rtol=5e-2),
-              verbose=false,
+              minibatch_size=nothing,
               # termination conditions
               maxiters::Int=100, #: The maximum number of iterations to perform. Defaults to 100.
               maxtime::Union{Nothing, <:Real}=nothing,# : The maximum time for solving the nonlinear system of equations. Defaults to nothing which means no time limit. Note that setting a time limit does have a small overhead.
@@ -96,8 +96,8 @@ function ksvd(Y::AbstractMatrix{T}, n_atoms::Int, max_nnz=n_atoms÷10;
               nnz_per_col_target::Number=0.0,
               # tracing options
               show_trace::Bool=false,
-              # store_trace::Bool,
               callback_fn::Union{Nothing, Function}=nothing,
+              verbose=false,
               ) where T
     timer = TimerOutput()
     emb_dim, n_samples = size(Y)
@@ -107,7 +107,9 @@ function ksvd(Y::AbstractMatrix{T}, n_atoms::Int, max_nnz=n_atoms÷10;
         D = init_dictionary(T, emb_dim, n_atoms)  # size(D) == (n, K)
         @assert all(≈(1.0), norm.(eachcol(D)))
     end
-    X = sparse(zeros(T, 0, 0))  # to assign to later
+    X = sparse_coding(sparse_coding_method, Y, D; timer)
+
+    Y_ = !isnothing(minibatch_size) ? similar(Y, size(Y, 1), minibatch_size) : similar(Y, 0, 0)
 
     # progressbar = Progress(maxiter)
     maybe_init_buffers!(ksvd_update_method, emb_dim, n_atoms, n_samples)
@@ -127,10 +129,18 @@ function ksvd(Y::AbstractMatrix{T}, n_atoms::Int, max_nnz=n_atoms÷10;
     termination_condition = :nothing
     tic = time()
     for iter in 1:maxiters
+        verbose && @info "Starting svd"
+        (Y_, X_) = if isnothing(minibatch_size)
+            minibatch_indices = sort(shuffle(axes(Y, 2))[1:minibatch_size])
+            Y_ .= Y[:, minibatch_indices]
+            X_ = X[:, minibatch_indices]
+            (Y_, X_)
+        else
+            (Y, X)
+        end
+        ksvd_update(ksvd_update_method, Y_, D, X_; timer)
         verbose && @info "Starting sparse coding"
         X = sparse_coding(sparse_coding_method, Y, D; timer)
-        verbose && @info "Starting svd"
-        D, X = ksvd_update(ksvd_update_method, Y, D, X; timer)
 
         # put a task to compute the trace / termination conditions.
         push!(trace_channel, (iter, copy(D), copy(X)))