-
Notifications
You must be signed in to change notification settings - Fork 0
Expand file tree
/
Copy pathmachine_learning.html
More file actions
260 lines (242 loc) · 13.7 KB
/
machine_learning.html
File metadata and controls
260 lines (242 loc) · 13.7 KB
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
<!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>
— the
probability of wildfire burning in a specific location — 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>© 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>