University of Utah CS 5140/6140 Data Mining
Professor Jeff Phillips · University of Utah · 10 lectures with notes
Students in this class: ask your lecturer for the class code, and these lectures will already be in your library when you sign up.
UUtah Data Mining | Fall 2026 | L13 : Streaming Freq Apx
Misra–Gries and Count-Min Sketch estimate frequent items in a stream with small memory and additive error guarantees.
UUtah Data Mining | Fall 2026 | L12: Steaming & Sampling
Reservoir and priority sampling maintain useful uniform or weighted samples from a stream using limited memory.
UUtah Data Mining | Fall 2026 | Choosing k (#clusters)
Use plots, elbow curves, silhouette scores, and—when a likelihood model is justified—BIC to choose a useful, not universally correct, number of clusters.
UUtah Data Mining | Fall 2026 | L10: Spectral Clustering
Spectral clustering uses normalized graph Laplacians and their low-eigenvalue embeddings to find balanced graph partitions.
UUtah Fall 2026 | Data Mining | L9 - k-Means and friends
Lloyd’s algorithm minimizes squared-Euclidean k-means cost locally; k-means++ makes its initialization substantially more reliable.
UUtah Data Mining | Fall 2026 | L8: Hierarchical Agglomerative Clustering
HAC linkage choices determine whether clustering favors compact groups, distant boundaries, or connected shapes.
UUtah Data Mining | Fall 2026 | L7 - LSH & Distribution Dist
LSH uses randomized hash collisions and banding to retrieve approximate neighbors, while distribution distances encode different modeling assumptions.
UUtah F2026 | Data Mining | L6 - Similarities
Jaccard similarity compares sets of text shingles, while MinHash estimates that similarity through randomized hash collisions.
UUtah Fall 2026 | Data Mining | L5 NN Search
HNSW combines layered neighbor graphs and beam search to make approximate nearest-neighbor retrieval practical in high-dimensional vector databases.
UUtah Fall 2026 | Data Mining | L4 - Metric Distances
Choosing a distance function defines similarity in data mining and can change which points an algorithm treats as neighbors.