Fix README

Author: david-baek

README.md

@@ -7,7 +7,8 @@ This is the GitHub repository for the paper "Harmonic Loss Trains Interpretable
 ## What is Harmonic Loss?
 - Harmonic logit $d_i$ is defined as the $l_2$ distance between the weight vector $\mathbf{w}_i$ and the input (query) $\mathbf{x}$:  $d_i = \|\mathbf{w}_i - \mathbf{x}\|_2$.
 
-- The probability $p_i$ is computed using the harmonic max function: <img src="./figures/eq_harmax.png" alt="Description" style="width: 50%; display: block; margin: auto;">
+- The probability $p_i$ is computed using the harmonic max function:
+![Harmonic Max]("./figures/eq_harmax.png")
  where $n$ is the **harmonic exponent**—a hyperparameter that controls the heavy-tailedness of the probability distribution.
 
 - Harmonic Loss achieves (1) **nonlinear separability**, (2)  **fast convergence**, (3) **scale invariance**, (4) **interpretability by design**, properties that are not available in cross-entropy loss.