<!DOCTYPE html><html><head>
<meta http-equiv="Content-Type" content="text/html; charset=utf-8">
  </head>
  <body>
    <div class="moz-forward-container">
      <div>
        <div dir="ltr">
          <div>Dear all,</div>
          <div><br>
          </div>
          <div>I have been asked to forward the below talk announcement
            by MSP alumnus Bruno. Personally I'm looking forward to it!<br>
          </div>
          <div><br>
          </div>
          <div>Best wishes,<br>
            Fred<br>
          </div>
          <div>
            <br>
            <div class="gmail_quote gmail_quote_container">
              <div dir="ltr" class="gmail_attr">---------- Forwarded
                message ---------<br>
                From: <strong class="gmail_sendername" dir="auto">Neil
                  Ghani</strong> <span dir="auto">
                  &lt;<a href="mailto:neil@kodamai.com" moz-do-not-send="true" class="moz-txt-link-freetext">neil@kodamai.com</a>&gt;</span><br>
                Date: Wed, Nov 12, 2025 at 2:06 PM<br>
                Subject: Fwd: Talk<br>
                To: Riu Sakamoto &lt;<a href="mailto:riu@kodamai.com" moz-do-not-send="true" class="moz-txt-link-freetext">riu@kodamai.com</a>&gt;<br>
              </div>
              <br>
              <br>
              <div dir="ltr">
                <div class="gmail_quote">
                  <div dir="ltr">
                    <div>Dear All</div>
                    <div><br>
                    </div>
                    <div>Kodamai is hosting the following talk this
                      Friday 14 November at 4pm UK time in the general
                      area of Category Theory and AI. If you are
                      interested, please contact
                      <a href="mailto:neil@kodamai.com" target="_blank" moz-do-not-send="true" class="moz-txt-link-freetext">neil@kodamai.com</a>
                      for the online url
                      <p><strong>Speaker: Bruno Gavranovic</strong></p>
                      <p><strong>TensorType: Implementing and extending
                          Deep Learning with Types</strong></p>
                      <p>Category theory has seen a rise in applications
                        in deep learning, bringing with it ideas and
                        tools from functional programming and dependent
                        type theory. However, implementing these ideas
                        in practice faces significant friction because
                        successful neural network frameworks only exist
                        in dynamically typed languages such as Python.
                        Attempts to implement these ideas in
                        statically-typed languages have struggled with
                        expressiveness and ergonomics, typically only
                        replicating what exists without imagining what
                        can be.</p>
                      <p>In this talk, I will introduce TensorType, a
                        tensor processing framework implemented in Idris
                        2 aiming to demonstrate that ergonomics and
                        rigor need not be at odds. TensorType provides
                        tensor operations checked at compile-time while
                        enabling fundamentally new capabilities: tensors
                        that branch and recurse instead of being
                        confined to rectangular shapes. I will walk
                        through the design choices behind TensorType,
                        showcase how containers power its core
                        abstractions, and explore strange new worlds it
                        enables.<br>
                      </p>
                    </div>
                  </div>
                  <div dir="ltr" class="gmail_signature">
                  </div>
                </div>
              </div>
            </div>
          </div>
        </div>
      </div>
    </div>
  </body>
</html>