<div dir="ltr">Thanks for the references, I've briefly looked at them and am looking forward to perusing them more closely. The interpretation of ML is a big thing of course. The machine gives you the results and it would sometimes be nice if it can be accompanied by some explanation of how it achieved it.<br><br>What you describe about your work sounds very interesting indeed. My gut feel is that, at least with the current generation of AI, human ingenuity and judgement would be far superior to AI in recognising similarities in the "hand" of the artisans involved. But AI could well be a powerful tool in assisting the humans?  </div><br><div class="gmail_quote"><div dir="ltr" class="gmail_attr">On Mon, 9 Jan 2023 at 18:50, Steve Smith <<a href="mailto:sasmyth@swcp.com">sasmyth@swcp.com</a>> wrote:<br></div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex">
  
    
  
  <div>
    <p>I was hoping this article would have more meat to it, but the
      main point seems highly relevant to a practical/introductory
      workshop such as the one you are developing:<br>
    </p>
    <p><a href="https://dl.acm.org/doi/10.1145/3544903.3544905" target="_blank">The Need
        for Interpretable Features: Taxonomy and Motivation <br>
      </a></p>
    <p><a href="https://scitechdaily.com/mit-taxonomy-helps-build-explainability-into-the-components-of-machine-learning-models/" target="_blank">https://scitechdaily.com/mit-taxonomy-helps-build-explainability-into-the-components-of-machine-learning-models/</a></p>
    <p>My limited experience with ML is that the convo/invo-lutions that
      developers go through to make their learning models work well tend
      to obscure the interpretability of the results.   Some of this
      seems unavoidable (inevitable?), particularly the highly technical
      reserved terminology that often exists in a given domain, but this
      team purports to help provide guidelines for minimizing the
      consequences...</p>
    <p>My main experience with this domain involved an attempt to find
      distance measures between the flake patterns in 3D scanned
      archaeological artifacts, starting with lithics but also aspiring
      to work with pottery and textiles...  essentially trying to
      recognize similarities in the "hand" of the artisans involved.<br>
    </p>
    <p><br>
    </p>
    <div>On 1/9/23 1:29 AM, Pieter Steenekamp
      wrote:<br>
    </div>
    <blockquote type="cite">
      
      <div dir="ltr"><a class="gmail_plusreply" id="m_8371091482889110803plusReplyChip-0">@Russ,  </a>
        <div><a class="gmail_plusreply"><br>
          </a>a) I appreciate the suggestion to include a simple neural
          network that can make predictions based on inputs and be
          trained using steepest descent to optimize its weights. This
          would be a valuable addition to my training material, as it
          provides a foundational understanding of how neural networks
          work. <br>
          <br>
          b) My focus is on providing practical training for
          professionals with deep domain knowledge but limited software
          experience, who are looking to utilize deep learning as a tool
          in their respective fields. In contrast, it seems that your
          focus is on understanding the inner workings of deep learning.
          Both approaches have their own merits, and it is important to
          cater to the needs and goals of different learners.<br>
          <br>
          Pieter</div>
      </div>
      <br>
      <div class="gmail_quote">
        <div dir="ltr" class="gmail_attr">On Sun, 8 Jan 2023 at 21:20,
          Marcus Daniels <<a href="mailto:marcus@snoutfarm.com" target="_blank">marcus@snoutfarm.com</a>>
          wrote:<br>
        </div>
        <blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex">
          <div dir="auto">
            The main defects of both R and Python are a lack of a typing
            system and high performance compilation.  I find R still
            follows (is used by) the statistics research community more
            than Python.   Common Lisp was always better than either.<br>
            <br>
            <div dir="ltr">Sent from my iPhone</div>
            <div dir="ltr"><br>
              <blockquote type="cite">On Jan 8, 2023, at 11:03 AM, Russ
                Abbott <<a href="mailto:russ.abbott@gmail.com" target="_blank">russ.abbott@gmail.com</a>>
                wrote:<br>
                <br>
              </blockquote>
            </div>
            <blockquote type="cite">
              <div dir="ltr">
                <div dir="ltr">
                  <div class="gmail_default" style="font-family:arial,helvetica,sans-serif;font-size:small;color:rgb(0,0,0)">As
                    indicated in my original reply, my interest in this
                    project grows from my relative ignorance of Deep
                    Learning. My career has focussed exclusively on
                    symbolic computing. I've worked with and taught (a)
                    functional programming, logic programming, and
                    related issues in advanced Python; (b) complex
                    systems, agent-based modeling, genetic algorithms,
                    and related evolutionary processes, (c) a bit of
                    constraint programming, especially in MiniZinc, and
                    (d) reinforcement learning as Q-learning, which is
                    reinforcement learning without neural nets. I've
                    always avoided neural nets--and more generally
                    numerical programming of any sort. </div>
                  <div class="gmail_default" style="font-family:arial,helvetica,sans-serif;font-size:small;color:rgb(0,0,0)"><br>
                  </div>
                  <div class="gmail_default" style="font-family:arial,helvetica,sans-serif;font-size:small;color:rgb(0,0,0)">Deep
                    learning has produced so many impressive results
                    that I've decided to devote much of my retirement
                    life to learning about it. I retired at the end of
                    Spring 2022 and (after a break) am now devoting much
                    of my time to learning more about Deep Neural Nets.
                    So far, I've dipped my brain into it at various
                    points. I think I've learned a fair amount. For
                    example, </div>
                  <div class="gmail_default" style="font-family:arial,helvetica,sans-serif;font-size:small;color:rgb(0,0,0)">
                    <ul>
                      <li>I now know how to build a neural net (NN) that
                        adds two numbers using a single layer with a
                        single neuron. It's really quite simple and is,
                        I think, a beautiful example of how NNs work. If
                        I were to teach an intro to NNs I'd start with
                        this.</li>
                      <li>I've gone through the Kaggle Deep Learning
                        sequence mentioned earlier. </li>
                      <li>I found a paper that shows how you can
                        approximate any differentiable function to any
                        degree of accuracy with a single-layer NN. (This
                        is a very nice result, although I believe it's
                        not used explicitly in building serious Deep NN
                        systems.)</li>
                      <li>From what I've seen so far, most serious DNNs
                        are built using Keras rather than PyTorch.</li>
                      <li>I've looked at Jeremy Howard's <a href="http://fast.ai" target="_blank">fast.ai</a> material. I
                        was going to go through the course but stopped
                        when I found that it uses PyTorch. Also, it
                        seems to be built on
                        <a href="http://fast.ai" target="_blank">fast.ai</a> libraries
                        that do a lot of the work for you without
                        explanation.  And it seems to focus almost
                        exclusively on Convolutional NNs. </li>
                      <li>My impression of DNNs is that to a great
                        extent they are <i>ad hoc</i>. There is no good
                        way to determine the best architecture to use
                        for a given problem. By architecture, I mean the
                        number of layers, the number of neurons in each
                        layer, the types of layers, the activation
                        functions to use, etc. </li>
                      <li>All DNNs that I've seen use Python as code
                        glue rather than R or some other language. I
                        like Python--so I'm pleased with that.</li>
                      <li>To build serious NNs one should learn the
                        Python libraries Numpy (array manipulation) and
                        Pandas (data processing). Numpy especially seems
                        to be used for virtually all DNNs that I've
                        seen. </li>
                      <li>Keras and probably PyTorch include a number of
                        special-purpose neurons and layers that can be
                        included in one's DNN. These include: a DropOut
                        layer, LSTM (short-long-term memory) neurons,
                        convolutional layers, recurrent neural net
                        layers (RNN), and more recently transformers,
                        which get credit for ChatGPT and related
                        programs. My impression is that these
                        special-purpose layers are
                        <i>ad hoc</i> in the same sense that functions
                        or libraries that one finds useful in a
                        programming language are <i>ad hoc</i>. They
                        have been very important for the success of
                        DNNs, but they came into existence because
                        people invented them in the same way that people
                        invented useful functions and libraries. </li>
                      <li>NN libraries also include a menagerie of
                        activation functions. An activation function
                        acts as the final control on the output of a
                        layer. Different activation functions are used
                        for different purposes. To be successful in
                        building a DNN, one must understand what those
                        activation functions do for you and which ones
                        to use. </li>
                      <li>I'm especially interested in DNNs that use
                        reinforcement learning. That's because the first
                        DNN work that impressed me was DeepMind's DNNs
                        that learned to play Atari games--and then Go,
                        etc. An important advantage of Reinforcement
                        Learning (RL) is that it doesn't depend on
                        mountains of labeled data. </li>
                      <li>I find RL systems more interesting than image
                        recognition systems. One of the striking
                        features of many image recognition systems is
                        that they can be thrown off by changing a small
                        number of pixels in an image. The changed image
                        would look to a human observer just like the
                        original, but it might fool a trained NN into
                        labeling the image as a banana rather than, say,
                        an automobile, which is what it really is. To
                        address this problem people have developed
                        Generative Adversarial Networks (GANs) which
                        attempt to find such weaknesses in a neural net
                        during training and then to train the NN not to
                        have those weaknesses. This is a fascinating
                        result, but as far as I can tell, it mainly
                        shows how fragile some NNs are and doesn't add
                        much conceptual depth to one's understanding of
                        how NNs work. </li>
                    </ul>
                    <div>I'm impressed with this list of things I sort
                      of know. If you had asked me before I started
                      writing this email I wouldn't have thought I had
                      learned as much as I have. Even so, I feel like I
                      don't understand much of it beyond a superficial
                      level. </div>
                    <div><br>
                    </div>
                    <div class="gmail_default">
                      <div>So far I've done all my exploration using
                        Google's Colab (Google's Python notebook
                        implementation) and Kaggle's similar Python
                        notebook implementation. (I prefer Colab to
                        Kaggle.) Using either one, it's super nice not
                        to have to download and install anything!</div>
                      <div><br>
                      </div>
                    </div>
                    <div style="color:rgb(34,34,34);font-family:Arial,Helvetica,sans-serif">
                      <div dir="ltr">
                        <div dir="ltr">
                          <div>
                            <div dir="ltr">
                              <div>
                                <div dir="ltr">
                                  <div>
                                    <div dir="ltr">
                                      <div>
                                        <div dir="ltr">
                                          <div>
                                            <div dir="ltr">
                                              <div>
                                                <div dir="ltr">
                                                  <div dir="ltr">
                                                    <div dir="ltr">
                                                      <div dir="ltr">
                                                        <div dir="ltr">
                                                          <div dir="ltr">
                                                          <div dir="ltr">
                                                          <div dir="ltr">
                                                          <div dir="ltr">
                                                          <div dir="ltr">
                                                          </div>
                                                          </div>
                                                          </div>
                                                          </div>
                                                          </div>
                                                        </div>
                                                      </div>
                                                    </div>
                                                  </div>
                                                </div>
                                              </div>
                                            </div>
                                          </div>
                                        </div>
                                      </div>
                                    </div>
                                  </div>
                                </div>
                              </div>
                            </div>
                          </div>
                        </div>
                      </div>
                    </div>
                    <div>I'm continuing my journey to learn more about
                      DNNs. I'd be happy to have company and to help
                      develop materials to teach about DNNs. (Developing
                      teaching materials always helps me learn the
                      subject being covered.)</div>
                    <div><br>
                    </div>
                  </div>
                  <div>
                    <div dir="ltr">
                      <div dir="ltr">
                        <div>
                          <div dir="ltr">
                            <div>
                              <div dir="ltr">
                                <div>
                                  <div dir="ltr">
                                    <div>
                                      <div dir="ltr">
                                        <div>
                                          <div dir="ltr">
                                            <div>
                                              <div dir="ltr">
                                                <div dir="ltr">
                                                  <div dir="ltr">
                                                    <div dir="ltr">
                                                      <div dir="ltr">
                                                        <div dir="ltr">
                                                          <div dir="ltr">
                                                          <div dir="ltr">
                                                          <div dir="ltr">
                                                          <div dir="ltr">
                                                          <div dir="ltr">
                                                          <div dir="ltr"><span></span>--
                                                          Russ Abbott  
                                                                       
                                                                       
                                                                  <br>
                                                          Professor
                                                          Emeritus,
                                                          Computer
                                                          Science<br>
                                                          California
                                                          State
                                                          University,
                                                          Los Angeles<br>
                                                          </div>
                                                          </div>
                                                          </div>
                                                          </div>
                                                          </div>
                                                          </div>
                                                        </div>
                                                      </div>
                                                    </div>
                                                  </div>
                                                </div>
                                              </div>
                                            </div>
                                          </div>
                                        </div>
                                      </div>
                                    </div>
                                  </div>
                                </div>
                              </div>
                            </div>
                          </div>
                        </div>
                      </div>
                    </div>
                  </div>
                  <br>
                </div>
                <br>
                <div class="gmail_quote">
                  <div dir="ltr" class="gmail_attr">On Sun, Jan 8, 2023
                    at 1:48 AM glen <<a href="mailto:gepropella@gmail.com" target="_blank">gepropella@gmail.com</a>>
                    wrote:<br>
                  </div>
                  <blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex">
                    Yes, the money/expertise bar is still pretty high.
                    But TANSTAAFL still applies. And the overwhelming
                    evidence is coming in that specific models do better
                    than those trained up on diverse data sets, "better"
                    meaning less prone to subtle bullsh¡t. What I find
                    fascinating is tools like OpenAI *facilitate*
                    trespassing. We have a wonderful bloom of
                    non-experts claiming they understand things like
                    "deep learning". But do they? An old internet meme
                    is brought to mind: "Do you even Linear Algebra,
                    bro?" >8^D<br>
                    <br>
                    On 1/8/23 01:06, Jochen Fromm wrote:<br>
                    > I have finished a number of Coursera courses
                    recently, including "Deep Learning & Neural
                    Networks with Keras" which was ok but not great. The
                    problems with deep learning are<br>
                    > <br>
                    > * to achieve impressive results like chatGPT
                    from OpenAi or LaMDA from Goggle you need to spend
                    millions on hardware<br>
                    > * only big organisations can afford to create
                    such expensive models<br>
                    > * the resulting network is s black box and it
                    is unclear why it works the way it does<br>
                    > <br>
                    > In the end it is just the same old back
                    propagation that has been known for decades, just on
                    more computers and trained on more data. Peter
                    Norvig calls it "The unreasonable effectiveness of
                    data"<br>
                    > <a href="https://research.google.com/pubs/archive/35179.pdf" rel="noreferrer" target="_blank">
                      https://research.google.com/pubs/archive/35179.pdf</a><br>
                    > <br>
                    > -J.<br>
                    > <br>
                    > <br>
                    > -------- Original message --------<br>
                    > From: Russ Abbott <<a href="mailto:russ.abbott@gmail.com" target="_blank">russ.abbott@gmail.com</a>><br>
                    > Date: 1/8/23 12:20 AM (GMT+01:00)<br>
                    > To: The Friday Morning Applied Complexity
                    Coffee Group <<a href="mailto:friam@redfish.com" target="_blank">friam@redfish.com</a>><br>
                    > Subject: Re: [FRIAM] Deep learning training
                    material<br>
                    > <br>
                    > Hi Pieter,<br>
                    > <br>
                    > A few comments.<br>
                    > <br>
                    >   * Much of the actual deep learning material
                    looks like it came from the Kaggle "Deep Learning
                    <<a href="https://www.kaggle.com/learn/intro-to-deep-learning" rel="noreferrer" target="_blank">https://www.kaggle.com/learn/intro-to-deep-learning</a>>"
                    sequence.<br>
                    >   * In my opinion, R is an ugly and /ad hoc/
                    language. I'd stick to Python.<br>
                    >   * More importantly, I would put the
                    How-to-use-Python stuff into a preliminary class.
                    Assume your audience knows how to use Python and
                    focus on Deep Learning. Given that, there is only a
                    minimal amount of information about Deep Learning in
                    the write-up. If I were to attend the workshop and
                    thought I would be learning about Deep Learning, I
                    would be disappointed--at least with what's covered
                    in the write-up.<br>
                    > <br>
                    >     I say this because I've been looking for a
                    good intro to Deep Learning. Even though I taught
                    Computer Science for many years, and am now retired,
                    I avoided Deep Learning because it was so
                    non-symbolic. My focus has always been on symbolic
                    computing. But Deep Learning has produced so many
                    extraordinarily impressive results, I decided I
                    should learn more about it. I haven't found any
                    really good material. If you are interested, I'd be
                    more than happy to work with you on developing some
                    introductory Deep Learning material. <br>
                    > <br>
                    > -- Russ Abbott<br>
                    > Professor Emeritus, Computer Science<br>
                    > California State University, Los Angeles<br>
                    > <br>
                    > <br>
                    > On Thu, Jan 5, 2023 at 11:31 AM Pieter
                    Steenekamp <<a href="mailto:pieters@randcontrols.co.za" target="_blank">pieters@randcontrols.co.za</a>
                    <mailto:<a href="mailto:pieters@randcontrols.co.za" target="_blank">pieters@randcontrols.co.za</a>>>
                    wrote:<br>
                    > <br>
                    >     Thanks to the kind support of OpenAI's
                    chatGPT, I am in the process of gathering materials
                    for a comprehensive and hands-on deep learning
                    workshop. Although it is still a work in progress, I
                    welcome any interested parties to take a look and
                    provide their valuable input. Thank you!<br>
                    > <br>
                    >     You can get it from:<br>
                    >     <a href="https://www.dropbox.com/s/eyx4iumb0439wlx/deep%20learning%20training%20rev%2005012023.zip?dl=0" rel="noreferrer" target="_blank">https://www.dropbox.com/s/eyx4iumb0439wlx/deep%20learning%20training%20rev%2005012023.zip?dl=0</a>
                    <<a href="https://www.dropbox.com/s/eyx4iumb0439wlx/deep%20learning%20training%20rev%2005012023.zip?dl=0" rel="noreferrer" target="_blank">https://www.dropbox.com/s/eyx4iumb0439wlx/deep%20learning%20training%20rev%2005012023.zip?dl=0</a>><br>
                    > <br>
                    <br>
                    -- <br>
                    ꙮ Mɥǝu ǝlǝdɥɐuʇs ɟᴉƃɥʇ' ʇɥǝ ƃɹɐss snɟɟǝɹs˙ ꙮ<br>
                    <br>
                    -. --- - / ...- .- .-.. .. -.. / -- --- .-. ... . /
                    -.-. --- -.. .<br>
                    FRIAM Applied Complexity Group listserv<br>
                    Fridays 9a-12p Friday St. Johns Cafe   /   Thursdays
                    9a-12p Zoom <a href="https://bit.ly/virtualfriam" rel="noreferrer" target="_blank">
                      https://bit.ly/virtualfriam</a><br>
                    to (un)subscribe <a href="http://redfish.com/mailman/listinfo/friam_redfish.com" rel="noreferrer" target="_blank">
http://redfish.com/mailman/listinfo/friam_redfish.com</a><br>
                    FRIAM-COMIC <a href="http://friam-comic.blogspot.com/" rel="noreferrer" target="_blank">
                      http://friam-comic.blogspot.com/</a><br>
                    archives:  5/2017 thru present <a href="https://redfish.com/pipermail/friam_redfish.com/" rel="noreferrer" target="_blank">
                      https://redfish.com/pipermail/friam_redfish.com/</a><br>
                      1/2003 thru 6/2021  <a href="http://friam.383.s1.nabble.com/" rel="noreferrer" target="_blank">
                      http://friam.383.s1.nabble.com/</a><br>
                  </blockquote>
                </div>
                <span>-. --- - / ...- .- .-.. .. -.. / -- --- .-. ... .
                  / -.-. --- -.. .</span><br>
                <span>FRIAM Applied Complexity Group listserv</span><br>
                <span>Fridays 9a-12p Friday St. Johns Cafe   /
                    Thursdays 9a-12p Zoom <a href="https://bit.ly/virtualfriam" target="_blank">https://bit.ly/virtualfriam</a></span><br>
                <span>to (un)subscribe <a href="http://redfish.com/mailman/listinfo/friam_redfish.com" target="_blank">http://redfish.com/mailman/listinfo/friam_redfish.com</a></span><br>
                <span>FRIAM-COMIC <a href="http://friam-comic.blogspot.com/" target="_blank">http://friam-comic.blogspot.com/</a></span><br>
                <span>archives:  5/2017 thru present <a href="https://redfish.com/pipermail/friam_redfish.com/" target="_blank">https://redfish.com/pipermail/friam_redfish.com/</a></span><br>
                <span> 1/2003 thru 6/2021  <a href="http://friam.383.s1.nabble.com/" target="_blank">http://friam.383.s1.nabble.com/</a></span><br>
              </div>
            </blockquote>
          </div>
          -. --- - / ...- .- .-.. .. -.. / -- --- .-. ... . / -.-. ---
          -.. .<br>
          FRIAM Applied Complexity Group listserv<br>
          Fridays 9a-12p Friday St. Johns Cafe   /   Thursdays 9a-12p
          Zoom <a href="https://bit.ly/virtualfriam" rel="noreferrer" target="_blank">https://bit.ly/virtualfriam</a><br>
          to (un)subscribe <a href="http://redfish.com/mailman/listinfo/friam_redfish.com" rel="noreferrer" target="_blank">http://redfish.com/mailman/listinfo/friam_redfish.com</a><br>
          FRIAM-COMIC <a href="http://friam-comic.blogspot.com/" rel="noreferrer" target="_blank">http://friam-comic.blogspot.com/</a><br>
          archives:  5/2017 thru present <a href="https://redfish.com/pipermail/friam_redfish.com/" rel="noreferrer" target="_blank">https://redfish.com/pipermail/friam_redfish.com/</a><br>
            1/2003 thru 6/2021  <a href="http://friam.383.s1.nabble.com/" rel="noreferrer" target="_blank">http://friam.383.s1.nabble.com/</a><br>
        </blockquote>
      </div>
      <br>
      <fieldset></fieldset>
      <pre>-. --- - / ...- .- .-.. .. -.. / -- --- .-. ... . / -.-. --- -.. .
FRIAM Applied Complexity Group listserv
Fridays 9a-12p Friday St. Johns Cafe   /   Thursdays 9a-12p Zoom <a href="https://bit.ly/virtualfriam" target="_blank">https://bit.ly/virtualfriam</a>
to (un)subscribe <a href="http://redfish.com/mailman/listinfo/friam_redfish.com" target="_blank">http://redfish.com/mailman/listinfo/friam_redfish.com</a>
FRIAM-COMIC <a href="http://friam-comic.blogspot.com/" target="_blank">http://friam-comic.blogspot.com/</a>
archives:  5/2017 thru present <a href="https://redfish.com/pipermail/friam_redfish.com/" target="_blank">https://redfish.com/pipermail/friam_redfish.com/</a>
  1/2003 thru 6/2021  <a href="http://friam.383.s1.nabble.com/" target="_blank">http://friam.383.s1.nabble.com/</a>
</pre>
    </blockquote>
  </div>

-. --- - / ...- .- .-.. .. -.. / -- --- .-. ... . / -.-. --- -.. .<br>
FRIAM Applied Complexity Group listserv<br>
Fridays 9a-12p Friday St. Johns Cafe   /   Thursdays 9a-12p Zoom <a href="https://bit.ly/virtualfriam" rel="noreferrer" target="_blank">https://bit.ly/virtualfriam</a><br>
to (un)subscribe <a href="http://redfish.com/mailman/listinfo/friam_redfish.com" rel="noreferrer" target="_blank">http://redfish.com/mailman/listinfo/friam_redfish.com</a><br>
FRIAM-COMIC <a href="http://friam-comic.blogspot.com/" rel="noreferrer" target="_blank">http://friam-comic.blogspot.com/</a><br>
archives:  5/2017 thru present <a href="https://redfish.com/pipermail/friam_redfish.com/" rel="noreferrer" target="_blank">https://redfish.com/pipermail/friam_redfish.com/</a><br>
  1/2003 thru 6/2021  <a href="http://friam.383.s1.nabble.com/" rel="noreferrer" target="_blank">http://friam.383.s1.nabble.com/</a><br>
</blockquote></div>