CLASS
FACULTY LECTURE HALL FACULTY β€’ CLASSROOM
Excellence in Pedagogy • Interactive Virtual Lecture Hall
1 PRES
REC
SG
sam gathigi Lecturer
Mic: Off (Click Start Mic)
1/4
Topic 4.2 • Optimization & Gradient Descent Formulations

Loss Surface Topology & Convex Convergence

sam gathigi • Department of Procurement & Supply Chain Strategy

CORE EXAM TOPIC
Midterm Weight: 25%
OBJECTIVE FUNCTION (MSE)
J(ΞΈ) = 1/2m βˆ‘ (h_ΞΈ(x^(i)) - y^(i))Β²
Sum of squared residuals over m training examples.
PARAMETER UPDATE RULE
ΞΈ_j := ΞΈ_j - Ξ± βˆ‚/βˆ‚ΞΈ_j J(ΞΈ)
Learning rate Ξ± tuned to 0.01 for stable convergence.
CONVERGENCE TOLERANCE
|| βˆ‡J(ΞΈ) ||β‚‚ ≤ 10⁻⁡
Stopping criterion verified across GPU cluster.
Loss Function J(ΞΈ) vs. Training Epochs (Live Telemetry) • Step 1 to 500
Minimum: Epoch 428 (0.0034 Loss)
SG
sam gathigi Lead Instructor • Podium
LECTURER PODIUM
sam gathigi
πŸ”Š Tap to Test Speaker