format

index.html

@@ -270,13 +270,9 @@ <h2 class="title is-3">Smooth Loss Landscapes</h2>
           <div class="item">
             <!-- Your image here -->
             <img style="margin-bottom:1cm;" src="static/images/grad_cropped.png" alt="By decomposing the problem into separate parts, B2B exhibits a smooth loss landscape"/>
-            <div class="columns is-centered has-text-centered">
-              <div class="column is-four-fifths">
                 Interestingly, decomposing the operator learning problem into separate parts leads to smoother loss
               landscapes compared to DeepONet, which leverages an end-to-end approach.
               This likely explains the faster convergence and better asymptotic performance of B2B.
-              </div>
-            </div>
           </div>
         </div>
       </div>
@@ -294,18 +290,14 @@ <h2 class="title is-3">Eigen Decomposition and Singular Value Decomposition</h2>
       <div class="content has-text-justified">
        <div class="item">
         <!-- Your image here -->
-        <img width="75%" style="margin-bottom:1cm;" src="static/images/svd_cropped_transparent.png" alt="With small modifications to our algorithm, we recover the familiar ED and SVD algorithms."/>
-        <div class="columns is-centered has-text-centered">
-          <div class="column is-four-fifths">
+        <img width="100%" style="margin-bottom:1cm;" src="static/images/svd_cropped_transparent.png" alt="With small modifications to our algorithm, we recover the familiar ED and SVD algorithms."/>
               B2B is rooted in classical functional analysis. As a result, with minimal modifications, B2B can recover the algorithms for eigen decomposition and singular value decomposition.
             While these approaches can only learn linear operators, they provide a means for analysing learned operators.
           </div>
         </div>
       </div>
     </div>
   </div>
-</div>
-</div>
 </section>