machine_learning.html

<!DOCTYPE HTML>
<!--
	Author: Fantine Huot
-->
<html>

<head>
    <title>Fantine Huot</title>
    <meta charset="utf-8" />
    <meta name="viewport" content="width=device-width, initial-scale=1, user-scalable=no" />
    <link rel="stylesheet" href="assets/css/main.css" />
    <noscript>
        <link rel="stylesheet" href="assets/css/noscript.css" />
    </noscript>
</head>

<body class="is-preload">

    <!-- Page Wrapper -->
    <div id="page-wrapper">

        <!-- Header -->
        <header id="header">
            <h1><a href="index.html">Fantine Huot</a></h1>
            <nav>
                <a href="#menu">Menu</a>
            </nav>
        </header>

        <!-- Menu -->
        <nav id="menu">
            <div class="inner">
                <h2>Menu</h2>
                <ul class="links">
                    <li><a href="index.html">Home</a></li>
                    <li><a href="machine_learning.html">AI Research</a></li>
                    <li><a href="science.html">Earth Sciences</a></li>
                    <li><a href="problem_solving.html">Problem-Solving</a></li>
                    <li><a href="about.html">About</a></li>
                </ul>
                <a href="#" class="close">Close</a>
            </div>
        </nav>

        <!-- Wrapper -->
        <section id="wrapper">
            <header>
                <div class="inner">
                    <h2>AI Research</h2>
                    <p>Pushing the boundaries of AI capabilities.</p>
                </div>
            </header>

            <!-- Content -->
            <div class="wrapper">
                <div class="inner">

                    <h3 class="major">Making language models more controllable and steerable</h3>
                    <div class="col-12"><span class="image left"><img src="images/gem.jpg" alt="" /></span></div>
                    <p>
                        While LLMs generate natural language well enough to create fluent text, it is still difficult to control the generation process, leading to irrelevant, repetitive, and hallucinated content. 
			At Google DeepMind, I have been working on planning for making language generation less opaque and more grounded. My work has shown that planning is particularly beneficial (a) <a href="https://aclanthology.org/2023.tacl-1.55/">when the input and output are long</a> (e.g., a book, multiple news articles, stories), 
			(b) <a href="https://aclanthology.org/2023.eacl-demo.13/">in a human-in-the-loop setting where a user works together with the model to edit text</a>; (c) <a href="https://arxiv.org/abs/2404.03381">in high-precision scenarios where the system’s output must be faithful to the input (e.g., with built-in citations)</a>; and (d) <a href="https://aclanthology.org/2024.eacl-long.131/">in cross-lingual generation and low-resource settings</a> in particular as a bridge between languages.
                    </p>
		    <p>
			    I got to be a small part of the large effort that led to Gemini, Google's largest and most capable AI model. Give it a try!
		    </p>
<!--                     <p>
                        <code>Python</code> <code>TensorFlow</code> <code>JAX</code>
                    </p> -->
                    <div class="row">
                        <div class="col-6 col-12-medium">
                            <ul class="actions small">
				<li><a href="https://gemini.google.com/" class="button primary small">Play with Gemini</a>
                                </li>
                                <li><a href="https://scholar.google.com/citations?user=79VvQLMAAAAJ&hl=en" class="button primary small">See publications</a>
                                </li>
                            </ul>
                        </div>
                    </div>
                </div>
                <div class="inner">

                    <h3 class="major">Estimating wildfire risk from remote-sensing data</h3>
                    <div class="col-12"><span class="image fit"><img src="images/fire.jpg" alt="" /></span></div>
                    <p>
                        Over the last few decades, wildfires have become a massive problem, with longer fire seasons and
                        larger
                        fires each year. Assessing the <b>fire likelihood</b>
                        &mdash; the
                        probability of wildfire burning in a specific location &mdash; is critical for forestry
                        management, disaster
                        preparedness, and early-warning systems.
                    </p>
                    <p>
                        Traditionally, we estimate wildfire likelihood by modeling fire behavior across simulations by
                        varying
                        parameters, which can be time-consuming and computationally-intensive. Instead of using
                        simulations, I
                        implemented and trained <b>deep neural networks</b> to predict the occurrence of wildfires from
                        remote-sensing data, using data from historical wildfires. I demonstrated that these models can
                        successfully
                        identify areas of high fire likelihood from aggregated data about vegetation, weather, and
                        topography.
                    </p>
                    <p>
                        I presented this work at the <a href="https://www.hadr.ai/home"><b>NeurIPS 2020</b>
                            Artificial
                            Intelligence for Humanitarian Assistance and
                            Disaster Response Workshop</a>.
                    </p>
                    <p>
                        <code>Python</code> <code>TensorFlow</code> <code>Earth Engine</code>
                    </p>
                    <div class="row">
                        <div class="col-6 col-12-medium">
                            <ul class="actions small">
                                <li><a href="https://www.hadr.ai/home" class="button primary small">NeurIPS workshop</a>
                                </li>
                                <li><a href="https://ieeexplore.ieee.org/abstract/document/9840400" class="button primary small">Get the
                                        paper</a></li>
                            </ul>
                        </div>
                    </div>
                </div>
                <div class="inner">
                    <h3 class="major">Training a neural network to find more earthquakes than ever before</h3>
                    <!-- <div class="col-12"><span class="image fit"><img src="images/tremor.png" alt="" /></span> </div> -->
                    <p>
                        When we think of earthquakes, we think of the ground shaking. But there exists an entirely
                        different type of
                        earthquake, much smaller in amplitude, similar to a repeating rumbling: These tiny repeating
                        earthquakes are
                        called <b>tectonic tremor</b>.
                    </p>
                    <p>
                        Since tectonic tremor repeats regularly, it is valuable for mapping the precise location of
                        faults. However,
                        it is near the noise level and thus <b>difficult to detect reliably</b>. It is usually detected
                        by looking
                        for a match using carefully crafted reference templates.
                    </p>
                    <p>
                        Using a catalog of more than <b>1 million tremor events</b> detected along the San Andreas Fault
                        for over 15
                        years, I created a novel approach to tremor detection with a convolutional neural network. I
                        demonstrate
                        that this methodology can detect tremor of very low signal amplitude, even below the background
                        noise level,
                        without prior templates.
                    </p>
                    <p>
                        <code>Python</code> <code>C++</code> <code>TensorFlow</code>
                    </p>
                    <div class="row">
                        <div class="col-6 col-12-medium">
                            <ul class="actions small">
                                <li><a href="https://library.seg.org/doi/abs/10.1190/segam2018-2998567.1"
                                        class="button primary small">Get
                                        the
                                        paper</a></li>
                            </ul>
                        </div>
                    </div>
                </div>

                    <!-- <h3 class="major">Vitae phasellus</h3>
									<p>Cras mattis ante fermentum, malesuada neque vitae, eleifend erat. Phasellus non pulvinar erat. Fusce tincidunt, nisl eget mattis egestas, purus ipsum consequat orci, sit amet lobortis lorem lacus in tellus. Sed ac elementum arcu. Quisque placerat auctor laoreet.</p>

									<section class="features">
										<article>
											<a href="#" class="image"><img src="images/pic04.jpg" alt="" /></a>
											<h3 class="major">Sed feugiat lorem</h3>
											<p>Lorem ipsum dolor sit amet, consectetur adipiscing vehicula id nulla dignissim dapibus ultrices.</p>
											<a href="#" class="special">Learn more</a>
										</article>
										<article>
											<a href="#" class="image"><img src="images/pic05.jpg" alt="" /></a>
											<h3 class="major">Nisl placerat</h3>
											<p>Lorem ipsum dolor sit amet, consectetur adipiscing vehicula id nulla dignissim dapibus ultrices.</p>
											<a href="#" class="special">Learn more</a>
										</article> -->
                    <!-- </section> -->
                </div>
            </div>

        </section>

        <!-- Footer -->
        <section id="footer">
            <div class="inner">
                <h2 class="major">Get in touch</h2>
                <p>I’m always open to interesting conversations and collaboration.</p>
                <form method="post" name="emailform" action="https://formspree.io/f/mjvpldab">
                    <div class="fields">
                        <div class="field">
                            <label for="name">Name</label>
                            <input type="text" name="name" id="name" />
                        </div>
                        <div class="field">
                            <label for="email">Email</label>
                            <input type="email" name="email" id="email" />
                        </div>
                        <div class="field">
                            <label for="message">Message</label>
                            <textarea name="message" id="message" rows="4"></textarea>
                        </div>
                    </div>
                    <ul class="actions">
                        <li><input type="submit" value="Send Message" /></li>
                    </ul>
                </form>
                <ul class="contact">
	                <li class="icon solid fa-home">Amsterdam, Netherlands</li>
	                <li class="icon solid fa-file"><a href="fantine_huot_resume_2024.pdf" target="_blank"
	                        rel="noopener noreferrer">My resume</a></li>
	                <li class="icon brands fa-google"><a
								href="https://scholar.google.com/citations?hl=en&user=79VvQLMAAAAJ&authuser=1" target="_blank" rel="noopener noreferrer">Google Scholar</a></li>
	                <li class="icon brands fa-linkedin"><a href="https://www.linkedin.com/in/fantine" target="_blank"
	                        rel="noopener noreferrer">linkedin.com/in/fantine</a>
			</li>
			    <!-- <li class="icon solid fa-phone">(000) 000-0000</li> -->
	                <!-- <li class="icon solid fa-envelope"><a href="#">information@untitled.tld</a></li> -->
	                <li class="icon brands fa-github"><a href="http://www.github.com/fantine" target="_blank"
	                        rel="noopener noreferrer">github.com/fantine</a>
	                </li>
	                <!-- <li class="icon brands fa-twitter"><a href="#">twitter.com/untitled-tld</a></li> -->
	                <!-- <li class="icon brands fa-facebook-f"><a href="#">facebook.com/untitled-tld</a></li> -->
	                <li class="icon brands fa-instagram"><a href="http://instagram.com/fantine.art" target="_blank"
	                        rel="noopener noreferrer">instagram.com/fantine.art</a></li>

                </ul>
                <ul class="copyright">
                    <li>&copy; 2024, Fantine Huot. All rights reserved.</li>
                </ul>
            </div>
        </section>

    </div>

    <!-- Scripts -->
    <script src="assets/js/jquery.min.js"></script>
    <script src="assets/js/jquery.scrollex.min.js"></script>
    <script src="assets/js/browser.min.js"></script>
    <script src="assets/js/breakpoints.min.js"></script>
    <script src="assets/js/util.js"></script>
    <script src="assets/js/main.js"></script>
    <script src="assets/js/gen_validatorv31.js"></script>
    <script language="JavaScript">
        // Code for validating the form
        // http://www.javascript-coder.com/html-form/javascript-form-validation.phtml
        var frmvalidator = new Validator("emailform");
        frmvalidator.addValidation("name", "req", "Please provide your name");
        frmvalidator.addValidation("email", "req", "Please provide your email");
        frmvalidator.addValidation("email", "email", "Please enter a valid email address"); 
    </script>

</body>

</html>