-
Notifications
You must be signed in to change notification settings - Fork 0
/
index.html
324 lines (275 loc) · 13.9 KB
/
index.html
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
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
<!DOCTYPE html>
<html>
<head>
<meta charset="utf-8">
<meta name="description"
content="DeLLMa: Decision Making Under Uncertainty with Large Language Models">
<meta name="keywords" content="Decision Theory, Large Lanaguge Models, GPT-4">
<meta name="viewport" content="width=device-width, initial-scale=1">
<title>DeLLMa: Decision Making Under Uncertainty with Large Language Models</title>
<!-- Google Tag Manager -->
<!-- <script>(function (w, d, s, l, i) {
w[l] = w[l] || []; w[l].push({
'gtm.start':
new Date().getTime(), event: 'gtm.js'
}); var f = d.getElementsByTagName(s)[0],
j = d.createElement(s), dl = l != 'dataLayer' ? '&l=' + l : ''; j.async = true; j.src =
'https://www.googletagmanager.com/gtm.js?id=' + i + dl; f.parentNode.insertBefore(j, f);
})(window, document, 'script', 'dataLayer', 'GTM-MFCT45H');</script> -->
<!-- End Google Tag Manager -->
<link href="https://fonts.googleapis.com/css?family=Google+Sans|Noto+Sans|Castoro"
rel="stylesheet">
<link rel="stylesheet" href="./static/css/bulma.min.css">
<link rel="stylesheet" href="./static/css/bulma-carousel.min.css">
<link rel="stylesheet" href="./static/css/bulma-slider.min.css">
<link rel="stylesheet" href="./static/css/fontawesome.all.min.css">
<link rel="stylesheet"
href="https://cdn.jsdelivr.net/gh/jpswalsh/academicons@1/css/academicons.min.css">
<link rel="stylesheet" href="./static/css/index.css">
<!-- <link rel="icon" href="./static/images/tifa.png"> -->
<script src="https://ajax.googleapis.com/ajax/libs/jquery/3.5.1/jquery.min.js"></script>
<script defer src="./static/js/fontawesome.all.min.js"></script>
<script src="./static/js/bulma-carousel.min.js"></script>
<script src="./static/js/bulma-slider.min.js"></script>
<script src="./static/js/index.js"></script>
</head>
<body>
<!-- Google Tag Manager (noscript) -->
<!-- <noscript><iframe src="https://www.googletagmanager.com/ns.html?id=GTM-MFCT45H" height="0" width="0"
style="display:none;visibility:hidden"></iframe></noscript> -->
<!-- End Google Tag Manager (noscript) -->
<nav class="navbar" role="navigation" aria-label="main navigation">
<div class="navbar-brand">
<a role="button" class="navbar-burger" aria-label="menu" aria-expanded="false">
<span aria-hidden="true"></span>
<span aria-hidden="true"></span>
<span aria-hidden="true"></span>
</a>
</div>
<div class="navbar-menu">
<div class="navbar-start" style="flex-grow: 1; justify-content: center;">
<div class="navbar-item has-dropdown is-hoverable">
<a class="navbar-link">
More Research
</a>
<div class="navbar-dropdown">
<a class="navbar-item" href="https://arxiv.org/abs/2305.00586">
How does GPT-2 compute greater-than?
</a>
<a class="navbar-item" href="https://arxiv.org/abs/2310.17086">
How Transformers learn in-context?
</a>
<a class="navbar-item" href="https://arxiv.org/abs/2310.07972">
How to interpret diffusion models?
</a>
<a class="navbar-item" href="https://arxiv.org/abs/2311.17946">
How to improve text-to-image alignment?
</a>
<a class="navbar-item" href="https://arxiv.org/abs/2312.06550">
LLM360: Towards Fully Transparent Open-Source LLMs
</a>
</div>
</div>
</div>
</div>
</nav>
<section class="hero">
<div class="hero-body">
<div class="container is-max-desktop">
<div class="columns is-centered">
<div class="column has-text-centered">
<h1 class="title is-1 publication-title"><span style="color: rgb(153, 27, 30);"><b>DeLLMa</b></span>: Decision Making Under Uncertainty with Large Language Models</h1>
<div class="is-size-5 publication-authors">
<span class="author-block">
<a href="https://ollieliu.com/">Ollie Liu</a><sup>1*</sup>,</span>
<span class="author-block">
<a href="https://sites.google.com/usc.edu/deqingfu">Deqing Fu</a><sup>1*</sup>,</span>
<span class="author-block">
<a href="https://dyogatama.github.io/">Dani Yogatama</a><sup>1</sup>,
</span>
<span class="author-block">
<a href="https://willieneis.github.io/">Willie Neiswanger</a><sup>1</sup>
</span>
</div>
<div class="is-size-5 publication-authors">
<span class="author-block"><sup>1 </sup>University of Southern California</span>
<span class="author-block"><sup>* </sup>Equal Contribution</span>
</div>
<div class="column has-text-centered">
<div class="publication-links">
<!-- PDF Link. -->
<span class="link-block">
<a href="https://arxiv.org/abs/2402.02392"
class="external-link button is-normal is-rounded is-dark">
<span class="icon">
<i class="fas fa-file-pdf"></i>
</span>
<span>Paper</span>
</a>
</span>
<!-- Code Link. -->
<span class="link-block">
<a href="https://github.com/DeLLMa/DeLLMa"
class="external-link button is-normal is-rounded is-dark">
<span class="icon">
<i class="fab fa-github"></i>
</span>
<span>Code</span>
</a>
</span>
</div>
</div>
</div>
</div>
</div>
</div>
</section>
<section class="hero teaser">
<div class="container is-max-desktop">
<div class="hero-body">
<h2 class="subtitle has-text-centered" style="max-width:870px;">
Classical decision theory has helped humans make rational decisions for decades. Can it do the same for Large Language Models (LLMs)?
<span class="dnerf" style="color: rgb(153, 27, 30);">DeLLMa</span> (pronouced dilemma) is a Decision-making Large Language Model assistant, which utilizes the power of <span style="color: red;"><b>Large Language Models</b></span> and <span style="color: red;"><b>Classical Decision Theory</b></span> to help LLMs make decisions under uncertainty.
</h2>
<div style="text-align: center;"><br/>
<img src="./static/images/dellma-overview-horizontal-new.png" alt="dellma teaser" width="800" class="center">
</div>
</div>
</div>
<div class="container is-max-desktop">
<!-- Abstract. -->
<div class="columns is-centered has-text-centered">
<div class="column is-four-fifths">
<h2 class="title is-3">Abstract</h2>
<div class="column content has-text-justified">
<p>
The potential of large language models (LLMs) as decision support tools is increasingly being explored in fields such as business, engineering, and medicine, which often face challenging tasks of decision-making under uncertainty. In this paper, we show that directly prompting LLMs on these types of decision-making problems can yield poor results, especially as the problem complexity increases. To aid in these tasks, we propose <span style="color: rgb(153, 27, 30);"><b>DeLLMa</b></span> (Decision-making Large Language Model assistant), a framework designed to enhance decision-making accuracy in uncertain environments. DeLLMa involves a multi-step reasoning procedure that integrates recent best practices in scaling inference-time reasoning, drawing upon principles from decision theory and utility theory, to provide an accurate and human-auditable decision-making process. We validate our procedure on multiple realistic decision-making environments, demonstrating that DeLLMa can consistently enhance the decision-making performance of leading language models, and achieve up to a 40% increase in accuracy over competing methods. Additionally, we show how performance improves when scaling compute at test time, and carry out human evaluations to benchmark components of DeLLMa.
</p>
</div>
</div>
</div>
</section>
<section class="section">
<div class="container is-max-desktop">
<div class="columns is-centered">
<div class="column is-four-fifths">
<h2 class="title is-3 has-text-centered">Project Overview</h2>
<!--<div class="content has-text-justified">-->
<div class="column content">
<p>
Given a decision query from a user, our framework DeLLMa (Decision-making LLM assistant) aims to perform accurate decision making under uncertainty via inference-time reasoning methods.
</p>
<p>
DeLLMa consists of four main steps:
<ol>
<li>Identify relevant <i>unknown states</i> based on the problem description and user goals.</li>
<li>Forecast the values of the unknown states given in-context information.</li>
<li>Elicit a <i>utility function</i> that aligns with the user’s goals.</li>
<li>Use this utility function to identify the decision that maximizes <i>expected utility</i>.</li>
</ol>
</p>
<p>
We show the full DeLLMa algorithm below:
</p>
<p>
<img src="./static/images/dellma-algorithm.png" alt="DeLLMa Algorithm" width="950">
</p>
<p>
We can illustrate the decision tree used internally by DeLLMa for decision making under
uncertainty. In the following figure, we show this decision tree for our agriculture
planning environment (described below). DeLLMa uses these types of decision trees to
compute and maximize the expected utility of each available action.
</p>
<p>
<img src="./static/images/decision-tree.png" alt="agriculture decision tree" width="950">
</p>
<h3 class="title is-3 has-text-centered">Some Experimental Results</h3>
<p>
We illustrate DeLLMa below on two decision making under uncertainty problems:
agriculture planning (<b>Agriculture</b>) and finance investing (<b>Stocks</b>).
Both problems involve sizable degrees of uncertainty from diverse sources, and are
representative of different data modalities (natural language and tabular) involved in
decision making
</p>
<p>
First, we show results on the Agriculture environment. We collect bi-annual reports
published by the United States Department of Agriculture (USDA) that provide analysis of
supply-and-demand conditions in the U.S. fruit markets. To emulate real-life farming
timelines, we use the report published in September 2021 as context for planning the
forthcoming agricultural year. We additionally supplement these natural language
contexts with USDA issued price and yield statistics in California.
</p>
<p>
We define the utility of planting a fruit as its price × yield reported in the
forthcoming year. We identify 7 fruits — apple, avocado, grape, grapefruit, lemon,
peach, and pear— that are both studied in the September 2021 report, and endowed with
these statistics in 2021 and 2022. We create decision making problems by enumerating all
possible combinations of availble fruits. For each decision-making instance, we use
related sections of the USDA report and current-year price and yield statistics as
context. In the figure below, we show that all DeLLMa variants outperform baseline
methods; DeLLMa-Pairs is the best, followed by Top1 and Naive. This result implies that
the full ranking of state-action pairs is useful for utility elicitation.
</p>
<p>
<img src="./static/images/scores_agriculture_acc_new.png" alt="agriculture scores" width="950", style="margin: 20px 0;">
</p>
<p>
Next, results on the Stocks environment. The action space A is limited to combinations
of 7 stocks: AMD, DIS, GME, GOOGL, META, NVDA and SPY. Unlike agriculture data where the
context C are collected through USDA reports, we collect historical stock prices as the
context for this problem. Each stock is presented with 24 monthly price in history. In
preventing possible data leakage and promoting LLMs to use their common-sense knowledge
in making decisions, when using gpt4-1106-preview as the LLM checkpoint, historical
price between December 2021 to November 2023 are provided as the context C. These
historical monthly prices are collected via Yahoo Finance by the authors.
</p>
<p>
The goal of the LLM agent is to choose which stock to invest on 2023-12-01 and sell on
the last trading day of that month (2023-12-29) so that the return is maximized. In the
figure below, we show that on average, DeLLMa-Top1 outperforms all baselines.
DeLLMa-Pairs is slightly worse than its Top1 counterpart, meaning that ranking
state-action pairs is still a challenging task.
</p>
<img src="./static/images/scores_stock_acc_new.png" alt="stock scores" width="950", style="margin: 20px 0;">
</div>
</div>
</div>
</div>
</section>
<section class="section" id="BibTeX">
<div class="container is-max-desktop content">
<h2 class="title">BibTeX</h2>
<pre><code>@misc{liu2024dellma,
title={{D}e{L}{L}{M}a: {D}ecision {M}aking {U}nder {U}ncertainty with {L}arge {L}anguage {M}odels},
author={Ollie Liu$^*$ and Deqing Fu$^*$ and Dani Yogatama and Willie Neiswanger},
year={2024},
eprint={2402.02392},
archivePrefix={arXiv},
primaryClass={cs.AI}
}</code></pre>
</div>
</section>
<footer class="footer">
<div class="container">
<div class="columns is-centered">
<div class="column is-8">
<div class="content">
<p>
This website is licensed under a <a rel="license"
href="http://creativecommons.org/licenses/by-sa/4.0/">Creative
Commons Attribution-ShareAlike 4.0 International License</a>.
</p>
<p>
This means you are free to borrow the <a
href="https://github.com/nerfies/nerfies.github.io">source code</a> of this website,
we just ask that you link back to this page in the footer.
Please remember to remove the analytics code included in the header of the website which
you do not want on your website.
</p>
</div>
</div>
</div>
</div>
</footer>
</body>
</html>