-
Notifications
You must be signed in to change notification settings - Fork 3
Expand file tree
/
Copy pathWorkshop_times.html
More file actions
258 lines (224 loc) · 9.42 KB
/
Workshop_times.html
File metadata and controls
258 lines (224 loc) · 9.42 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
<!DOCTYPE html>
<html>
<head>
<link rel="stylesheet" href="stylesheet.css">
<link rel="icon" href="./images/pacifier.png">
</head>
<body>
<div style="display:inline">
<img style="float: left; padding-right: 20px;" src="./images/pacifier.png" height="80">
<div class="master-title"> <b>Baby</b>LM Challenge </div>
<div class="subheader"> Sample-efficient pretraining on a developmentally plausible corpus </div>
</div>
<div id="navbar">
<h4> <a href="index.html"> Overview </a> • <a href="Workshop_times.html"> Workshop Schedule </a> • <a href="posters.html"> Posters </a> • <a href="guidelines.html"> Guidelines </a> • <a href="timeline.html"> Timeline</a> • <a href="faqs.html"> FAQs </a>• <a href="papers.html"> Previous papers </a> <hr> </h4>
</div>
2026 TBD
<!--
<section class="schedule" aria-labelledby="schedule-heading">
<h2 id="schedule-heading">Workshop Schedule 2025 · Saturday, November 8</h2>
<ul class="schedule-list">
<li><span class="time">09:00-09:15 </span><span class="title">Opening Remarks</span></li>
<li>
<span class="time">09:15-10:15 </span>
<span class="title">
BabyLM Challenge Orals
<span class="papers">
<div class="paper-entry">
<span class="paper-title">CLASS-IT: Conversational and Lecture-Aligned Small-Scale Instruction Tuning for BabyLMs</span>
<span class="paper-authors">Luca Capone, Alessandro Bondielli, Alessandro Lenci</span>
</div>
<div class="paper-entry">
<span class="paper-title">Masked Diffusion Language Models with Frequency-Informed Training</span>
<span class="paper-authors">Despoina Kosmopoulou, Efthymios Georgiou, Vaggelis Dorovatas, Georgios Paraskevopoulos, Alexandros Potamianos</span>
</div>
<div class="paper-entry">
<span class="paper-title">MoEP: Modular Expert Paths for Sample-Efficient Language Modeling</span>
<span class="paper-authors">Joonas Tapaninaho</span>
</div>
<div class="paper-entry">
<span class="paper-title">Mask and You Shall Receive: Optimizing Masked Language Modeling for Pre-training BabyLMs</span>
<span class="paper-authors">Lukas Edman, Alexander Fraser</span>
</div>
<div class="paper-entry">
<span class="paper-title"> Once Upon a Time: Interactive Learning for Storytelling with Small Language Models</span>
<span class="paper-authors">Jonas Mayer Martins, Ali Hamza Bashir, Muhammad Rehan Khalid, Lisa Beinborn</span>
</div>
</span>
</span>
</li>
<li><span class="time">10:15-11:00 </span><span class="title">Break</span></li>
<li><span class="time">11:00-12:00 </span><span class="title">Invited Talk 1: Hai Hu - Benchmarking Baby and Large Language Models in Chinese</span></li>
<div class="abstract static-abstract">
<div class="abstract-label">Abstract</div>
In this talk, I will first briefly overview the efforts in creating linguistically oriented benchmarks in Chinese. I will discuss the design and construction of benchmarks targeting the orthography, phonology, syntax, logic and semantics, pragmatics and world knowledge of modern and classical Chinese, by our lab and other teams in the field. The evaluations of baby and large language models show that current LLMs are very powerful, especially with the addition of “reasoning” abilities. However, certain linguistic blind spots remain, and further refinement of evaluation tasks and methodologies is needed. Next, I will discuss ongoing studies in understanding the learning mechanisms of monolingual and bilingual language models involving Chinese. Finally, I will point out what the LM community might learn from the language acquisition community.
</div>
<li><span class="time">12:00-13:30 </span><span class="title">Lunch</span></li>
<li><span class="time">13:30-15:00 </span><span class="title">Poster Session</span></li>
<li><span class="time">15:00-15:30 </span><span class="title">Break</span></li>
<li><span class="time">15:30-16:30 </span><span class="title">Invited Talk 2: Yohei Oseki - Small Language Models through Insights from Human Language Acquisition</span></li>
<div class="abstract static-abstract">
<div class="abstract-label">Abstract</div>
Large language models (LLMs) have achieved remarkable success, thanks to the rapid development of AI and machine/deep learning, and outperformed humans at various downstream tasks. However, those LLMs, despite their super-human performance, have been pointed out as not efficient in terms of training data, model parameters, and computational resources. In this talk, I propose small language models (SLMs) that efficiently learn natural language like humans, building on insights from human language acquisition. Specifically, SLMs are trained on developmentally plausible corpora like BabyLM Challenge via curriculum learning, batch learning, direct/indirect evidence, variation set, and critical period. The results suggest that inductive biases are essential to efficiently train SLMs, with scientific implications for human language acquisition, as well as engineering applications to edge AI and low-resource languages.
</div>
<li>
<span class="time">16:30-17:05 </span>
<span class="title">
BabyLM Workshop Orals
<span class="papers">
<div class="paper-entry">
<span class="paper-title">Teacher Demonstrations in a BabyLM’s Zone of Proximal Development for Contingent Multi-Turn Interaction</span>
<span class="paper-authors">Suchir Salhan, Hongyi Gu, Donya Rooein, Diana Galvan-Sosa, Gabrielle Gaudeau, Andrew Caines, Zheng Yuan, Paula Buttery</span>
</div>
<div class="paper-entry">
<span class="paper-title">Are BabyLMs Deaf to Gricean Maxims? A Pragmatic Evaluation of Sample-Efficient Language Models</span>
<span class="paper-authors">Raha Askari, Sina Zarrieß, Özge Alacam, Judith Sieker</span>
</div>
<div class="paper-entry">
<span class="paper-title">Looking to Learn: Token-wise Dynamic Gating for Low-Resource Vision-Language Modelling</span>
<span class="paper-authors">Bianca-Mihaela Ganescu, Suchir Salhan, Andrew Caines, Paula Buttery</span>
</div>
</span>
</span>
</li>
<li><span class="time">17:05-17:15 </span><span class="title">Awards and Closing Remarks</span></li>
</ul>
</section> -->
<style>
.schedule {
font-family: system-ui, sans-serif;
}
.schedule h2 {
font-size: 1.4rem;
font-weight: 600;
margin-bottom: 1rem;
color: #222;
}
.schedule-list {
list-style: none;
margin: 0;
padding: 0;
}
.schedule-list li {
display: grid;
grid-template-columns: auto 1fr;
align-items: baseline;
gap: 1rem;
padding: 0.01rem 0; /* tightened spacing */
border: none !important; /* kill inherited borders */
}
.schedule-list li * {
border: none !important; /* remove dotted or solid lines */
text-decoration: none !important; /* remove link-style underlines */
}
.time {
font-weight: 500;
color: #666;
text-align: left; /* was right */
font-variant-numeric: tabular-nums;
white-space: nowrap;
}
.paper-entry {
margin-bottom: 0.75rem; /* space between papers */
}
.paper-title {
display: block;
font-weight: 600;
color: #222;
}
.paper-authors {
display: block;
font-size: 0.9rem;
color: #555;
margin-left: 1rem; /* optional indent for visual hierarchy */
}
.title {
color: #222;
font-weight: 500;
line-height: 1.35;
}
/* small, inline list of papers/authors without changing layout */
.papers {
display: block;
margin-top: 1rem;
font-weight: 400;
font-size: 0.92rem;
color: #444;
}
.schedule-list li:nth-child(5),
.schedule-list li:nth-child(6) {
background-color: #fafafa;
}
.schedule-list li.talk {
display: grid;
grid-template-columns: auto 1fr;
gap: 1rem;
padding: 0.6rem 0;
border-top: 1px solid #eee !important;
}
.slot {
display: grid;
gap: 0.35rem;
}
.slot-top {
display: flex;
flex-wrap: wrap;
gap: 0.5rem 0.8rem;
align-items: baseline;
}
.slot-label {
font-size: 0.85rem;
font-weight: 600;
text-transform: uppercase;
letter-spacing: .02em;
color: #555;
background: #f3f4f6;
border-radius: 999px;
padding: 0.15rem 0.5rem;
}
.speaker {
font-weight: 500;
color: #333;
}
.talk-title {
font-weight: 650;
color: #111;
line-height: 1.35;
}
.abstract {
border-left: 3px solid #e5e7eb;
padding-left: 0.8rem;
}
.abstract > summary {
cursor: pointer;
font-weight: 500;
color: #444;
list-style: none;
}
.abstract > summary::-webkit-details-marker { display: none; }
.abstract > summary::after {
content: " ▾";
font-weight: 400;
}
.abstract[open] > summary::after { content: " ▴"; }
.abstract p {
margin: 0.4rem 0 0;
color: #333;
}
@media (max-width: 1200px) {
.schedule-list li.talk { grid-template-columns: 1fr; }
}
@media (max-width: 1200px) {
.schedule-list li {
grid-template-columns: 1fr;
gap: 0.2rem;
padding: 0.5rem 0;
}
.time { text-align: left; }
}
.static-abstract { border-left: 3px solid #e5e7eb; padding-left: .8rem; }
.abstract-label { font-weight: 500; color: #444; margin-bottom: .25rem; }
</style>
<div class="footer">
</div>
</body>