-
Notifications
You must be signed in to change notification settings - Fork 0
Expand file tree
/
Copy pathindex.html
More file actions
575 lines (531 loc) · 32.3 KB
/
index.html
File metadata and controls
575 lines (531 loc) · 32.3 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
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
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
<!DOCTYPE html>
<html lang="en" dir="auto">
<head>
<meta name="generator" content="Hugo 0.92.2" /><meta charset="utf-8">
<meta http-equiv="X-UA-Compatible" content="IE=edge">
<meta name="viewport" content="width=device-width, initial-scale=1, shrink-to-fit=no">
<meta name="robots" content="index, follow">
<title>Lil'Log</title>
<meta name="description" content="Document my learning notes.">
<meta name="author" content="Zhirui Zhang">
<link rel="canonical" href="https://jszzr.github.io/" />
<link crossorigin="anonymous" href="/assets/css/stylesheet.min.67a6fb6e33089cb29e856bcc95d7aa39f70049a42b123105531265a0d9f1258b.css" integrity="sha256-Z6b7bjMInLKehWvMldeqOfcASaQrEjEFUxJloNnxJYs=" rel="preload stylesheet" as="style">
<link rel="icon" href="https://jszzr.github.io/favicon_peach.ico">
<link rel="icon" type="image/png" sizes="16x16" href="https://jszzr.github.io/favicon-16x16.png">
<link rel="icon" type="image/png" sizes="32x32" href="https://jszzr.github.io/favicon-32x32.png">
<link rel="apple-touch-icon" href="https://jszzr.github.io/apple-touch-icon.png">
<link rel="mask-icon" href="https://jszzr.github.io/safari-pinned-tab.svg">
<meta name="theme-color" content="#2e2e33">
<meta name="msapplication-TileColor" content="#2e2e33">
<link rel="alternate" type="application/rss+xml" href="https://jszzr.github.io/index.xml">
<link rel="alternate" type="application/json" href="https://jszzr.github.io/index.json">
<noscript>
<style>
#theme-toggle,
.top-link {
display: none;
}
</style>
<style>
@media (prefers-color-scheme: dark) {
:root {
--theme: rgb(29, 30, 32);
--entry: rgb(46, 46, 51);
--primary: rgb(218, 218, 219);
--secondary: rgb(155, 156, 157);
--tertiary: rgb(65, 66, 68);
--content: rgb(196, 196, 197);
--hljs-bg: rgb(46, 46, 51);
--code-bg: rgb(55, 56, 62);
--border: rgb(51, 51, 51);
}
.list {
background: var(--theme);
}
.list:not(.dark)::-webkit-scrollbar-track {
background: 0 0;
}
.list:not(.dark)::-webkit-scrollbar-thumb {
border-color: var(--theme);
}
}
</style>
</noscript>
<script async src="https://www.googletagmanager.com/gtag/js?id=G-HFT45VFBX6"></script>
<script>
var doNotTrack = false;
if (!doNotTrack) {
window.dataLayer = window.dataLayer || [];
function gtag(){dataLayer.push(arguments);}
gtag('js', new Date());
gtag('config', 'G-HFT45VFBX6', { 'anonymize_ip': false });
}
</script>
<meta property="og:title" content="Lil'Log" />
<meta property="og:description" content="Document my learning notes." />
<meta property="og:type" content="website" />
<meta property="og:url" content="https://jszzr.github.io/" />
<meta name="twitter:card" content="summary"/>
<meta name="twitter:title" content="Lil'Log"/>
<meta name="twitter:description" content="Document my learning notes."/>
<script type="application/ld+json">
{
"@context": "https://schema.org",
"@type": "Organization",
"name": "Zhirui's Log",
"url": "https://jszzr.github.io/",
"description": "Document my learning notes.",
"thumbnailUrl": "https://jszzr.github.io/favicon_peach.ico",
"sameAs": [
"index.xml", "https://scholar.google.com/citations?user=dCa-pW8AAAAJ\u0026hl=en\u0026oi=ao", "https://github.com/jszzr", "https://twitter.com/jszzr/"
]
}
</script>
</head>
<body class="list" id="top">
<script>
if (localStorage.getItem("pref-theme") === "dark") {
document.body.classList.add('dark');
} else if (localStorage.getItem("pref-theme") === "light") {
document.body.classList.remove('dark')
} else if (window.matchMedia('(prefers-color-scheme: dark)').matches) {
document.body.classList.add('dark');
}
</script>
<script>
MathJax = {
tex: {
inlineMath: [['$', '$'], ['\\(', '\\)']],
displayMath: [['$$','$$'], ['\\[', '\\]']],
processEscapes: true,
processEnvironments: true
},
options: {
skipHtmlTags: ['script', 'noscript', 'style', 'textarea', 'pre']
}
};
window.addEventListener('load', (event) => {
document.querySelectorAll("mjx-container").forEach(function(x){
x.parentElement.classList += 'has-jax'})
});
</script>
<script src="https://polyfill.io/v3/polyfill.min.js?features=es6"></script>
<script type="text/javascript" id="MathJax-script" async
src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-mml-chtml.js"></script>
<header class="header">
<nav class="nav">
<div class="logo">
<a href="https://jszzr.github.io/" accesskey="h" title="Lil'Log (Alt + H)">Lil'Log</a>
<span class="logo-switches">
<button id="theme-toggle" accesskey="t" title="(Alt + T)">
<svg id="moon" xmlns="http://www.w3.org/2000/svg" width="24" height="24" viewBox="0 0 24 24"
fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round"
stroke-linejoin="round">
<path d="M21 12.79A9 9 0 1 1 11.21 3 7 7 0 0 0 21 12.79z"></path>
</svg>
<svg id="sun" xmlns="http://www.w3.org/2000/svg" width="24" height="24" viewBox="0 0 24 24"
fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round"
stroke-linejoin="round">
<circle cx="12" cy="12" r="5"></circle>
<line x1="12" y1="1" x2="12" y2="3"></line>
<line x1="12" y1="21" x2="12" y2="23"></line>
<line x1="4.22" y1="4.22" x2="5.64" y2="5.64"></line>
<line x1="18.36" y1="18.36" x2="19.78" y2="19.78"></line>
<line x1="1" y1="12" x2="3" y2="12"></line>
<line x1="21" y1="12" x2="23" y2="12"></line>
<line x1="4.22" y1="19.78" x2="5.64" y2="18.36"></line>
<line x1="18.36" y1="5.64" x2="19.78" y2="4.22"></line>
</svg>
</button>
</span>
</div>
<ul id="menu">
<li>
<a href="https://jszzr.github.io/" title="Posts">
<span class="active">Posts</span>
</a>
</li>
<li>
<a href="https://jszzr.github.io/archives" title="Archive">
<span>Archive</span>
</a>
</li>
<li>
<a href="https://jszzr.github.io/search/" title="Search (Alt + /)" accesskey=/>
<span>Search</span>
</a>
</li>
<li>
<a href="https://jszzr.github.io/tags/" title="Tags">
<span>Tags</span>
</a>
</li>
<li>
<a href="https://jszzr.github.io/faq" title="FAQ">
<span>FAQ</span>
</a>
</li>
<li>
<a href="https://www.emojisearch.app/" title="emojisearch.app">
<span>emojisearch.app</span>
</a>
</li>
</ul>
</nav>
</header>
<main class="main">
<article class="first-entry home-info">
<header class="entry-header">
<h1>👋 Welcome to Lil’Log</h1>
</header>
<section class="entry-content">
<p>Hi, this is Zhirui. I’m documenting my learning notes in this blog. Besides, I’m leading a team on AI Safety at OpenAI.</p>
</section>
<footer class="entry-footer">
<div class="social-icons">
<a href="index.xml" target="_blank" rel="noopener noreferrer me" title="Rss">
<svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2"
stroke-linecap="round" stroke-linejoin="round">
<path d="M4 11a9 9 0 0 1 9 9" />
<path d="M4 4a16 16 0 0 1 16 16" />
<circle cx="5" cy="19" r="1" />
</svg>
</a>
<a href="https://scholar.google.com/citations?user=dCa-pW8AAAAJ&hl=en&oi=ao" target="_blank" rel="noopener noreferrer me" title="Other">
<svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2"
stroke-linecap="round" stroke-linejoin="round">
<path d="M10 13a5 5 0 0 0 7.54.54l3-3a5 5 0 0 0-7.07-7.07l-1.72 1.71"></path>
<path d="M14 11a5 5 0 0 0-7.54-.54l-3 3a5 5 0 0 0 7.07 7.07l1.71-1.71"></path>
</svg>
</a>
<a href="https://github.com/jszzr" target="_blank" rel="noopener noreferrer me" title="Github">
<svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2"
stroke-linecap="round" stroke-linejoin="round">
<path
d="M9 19c-5 1.5-5-2.5-7-3m14 6v-3.87a3.37 3.37 0 0 0-.94-2.61c3.14-.35 6.44-1.54 6.44-7A5.44 5.44 0 0 0 20 4.77 5.07 5.07 0 0 0 19.91 1S18.73.65 16 2.48a13.38 13.38 0 0 0-7 0C6.27.65 5.09 1 5.09 1A5.07 5.07 0 0 0 5 4.77a5.44 5.44 0 0 0-1.5 3.78c0 5.42 3.3 6.61 6.44 7A3.37 3.37 0 0 0 9 18.13V22">
</path>
</svg>
</a>
<a href="https://twitter.com/jszzr/" target="_blank" rel="noopener noreferrer me" title="Twitter">
<svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2"
stroke-linecap="round" stroke-linejoin="round">
<path
d="M23 3a10.9 10.9 0 0 1-3.14 1.53 4.48 4.48 0 0 0-7.86 3v1A10.66 10.66 0 0 1 3 4s-4 9 5 13a11.64 11.64 0 0 1-7 2c9 5 20 0 20-11.5a4.5 4.5 0 0 0-.08-.83A7.72 7.72 0 0 0 23 3z">
</path>
</svg>
</a>
</div>
</footer>
</article>
<article class="post-entry">
<header class="entry-header">
<h2>LLM Powered Autonomous Agents
</h2>
</header>
<section class="entry-content">
<p>Building agents with LLM (large language model) as its core controller is a cool concept. Several proof-of-concepts demos, such as AutoGPT, GPT-Engineer and BabyAGI, serve as inspiring examples. The potentiality of LLM extends beyond generating well-written copies, stories, essays and programs; it can be framed as a powerful general problem solver.
Agent System Overview In a LLM-powered autonomous agent system, LLM functions as the agent’s brain, complemented by several key components:...</p>
</section>
<footer class="entry-footer"><span title='2023-06-23 00:00:00 +0000 UTC'>June 23, 2023</span> · 31 min · Zhirui Zhang</footer>
<a class="entry-link" aria-label="post link to LLM Powered Autonomous Agents" href="https://jszzr.github.io/posts/2023-06-23-agent/"></a>
</article>
<article class="post-entry">
<header class="entry-header">
<h2>Prompt Engineering
</h2>
</header>
<section class="entry-content">
<p>Prompt Engineering, also known as In-Context Prompting, refers to methods for how to communicate with LLM to steer its behavior for desired outcomes without updating the model weights. It is an empirical science and the effect of prompt engineering methods can vary a lot among models, thus requiring heavy experimentation and heuristics.
This post only focuses on prompt engineering for autoregressive language models, so nothing with Cloze tests, image generation or multimodality models....</p>
</section>
<footer class="entry-footer"><span title='2023-03-15 00:00:00 +0000 UTC'>March 15, 2023</span> · 21 min · Zhirui Zhang</footer>
<a class="entry-link" aria-label="post link to Prompt Engineering" href="https://jszzr.github.io/posts/2023-03-15-prompt-engineering/"></a>
</article>
<article class="post-entry">
<header class="entry-header">
<h2>The Transformer Family Version 2.0
</h2>
</header>
<section class="entry-content">
<p>Many new Transformer architecture improvements have been proposed since my last post on “The Transformer Family” about three years ago. Here I did a big refactoring and enrichment of that 2020 post — restructure the hierarchy of sections and improve many sections with more recent papers. Version 2.0 is a superset of the old version, about twice the length.
Notations Symbol Meaning $d$ The model size / hidden state dimension / positional encoding size....</p>
</section>
<footer class="entry-footer"><span title='2023-01-27 00:00:00 +0000 UTC'>January 27, 2023</span> · 45 min · Zhirui Zhang</footer>
<a class="entry-link" aria-label="post link to The Transformer Family Version 2.0" href="https://jszzr.github.io/posts/2023-01-27-the-transformer-family-v2/"></a>
</article>
<article class="post-entry">
<header class="entry-header">
<h2>Large Transformer Model Inference Optimization
</h2>
</header>
<section class="entry-content">
<p>[Updated on 2023-01-24: add a small section on Distillation.]
Large transformer models are mainstream nowadays, creating SoTA results for a variety of tasks. They are powerful but very expensive to train and use. The extremely high inference cost, in both time and memory, is a big bottleneck for adopting a powerful transformer for solving real-world tasks at scale.
Why is it hard to run inference for large transformer models? Besides the increasing size of SoTA models, there are two main factors contributing to the inference challenge (Pope et al....</p>
</section>
<footer class="entry-footer"><span title='2023-01-10 10:00:00 -0700 -0700'>January 10, 2023</span> · 31 min · Zhirui Zhang</footer>
<a class="entry-link" aria-label="post link to Large Transformer Model Inference Optimization" href="https://jszzr.github.io/posts/2023-01-10-inference-optimization/"></a>
</article>
<article class="post-entry">
<header class="entry-header">
<h2>Some Math behind Neural Tangent Kernel
</h2>
</header>
<section class="entry-content">
<p>Neural networks are well known to be over-parameterized and can often easily fit data with near-zero training loss with decent generalization performance on test dataset. Although all these parameters are initialized at random, the optimization process can consistently lead to similarly good outcomes. And this is true even when the number of model parameters exceeds the number of training data points.
Neural tangent kernel (NTK) (Jacot et al. 2018) is a kernel to explain the evolution of neural networks during training via gradient descent....</p>
</section>
<footer class="entry-footer"><span title='2022-09-08 10:00:00 -0700 PDT'>September 8, 2022</span> · 17 min · Zhirui Zhang</footer>
<a class="entry-link" aria-label="post link to Some Math behind Neural Tangent Kernel" href="https://jszzr.github.io/posts/2022-09-08-ntk/"></a>
</article>
<article class="post-entry">
<header class="entry-header">
<h2>Generalized Visual Language Models
</h2>
</header>
<section class="entry-content">
<p>Processing images to generate text, such as image captioning and visual question-answering, has been studied for years. Traditionally such systems rely on an object detection network as a vision encoder to capture visual features and then produce text via a text decoder. Given a large amount of existing literature, in this post, I would like to only focus on one approach for solving vision language tasks, which is to extend pre-trained generalized language models to be capable of consuming visual signals....</p>
</section>
<footer class="entry-footer"><span title='2022-06-09 15:10:30 -0700 PDT'>June 9, 2022</span> · 25 min · Zhirui Zhang</footer>
<a class="entry-link" aria-label="post link to Generalized Visual Language Models" href="https://jszzr.github.io/posts/2022-06-09-vlm/"></a>
</article>
<article class="post-entry">
<header class="entry-header">
<h2>Learning with not Enough Data Part 3: Data Generation
</h2>
</header>
<section class="entry-content">
<p>Here comes the Part 3 on learning with not enough data (Previous: Part 1 and Part 2). Let’s consider two approaches for generating synthetic data for training.
Augmented data. Given a set of existing training samples, we can apply a variety of augmentation, distortion and transformation to derive new data points without losing the key attributes. We have covered a bunch of augmentation methods on text and images in a previous post on contrastive learning....</p>
</section>
<footer class="entry-footer"><span title='2022-04-15 15:10:30 -0700 PDT'>April 15, 2022</span> · 28 min · Zhirui Zhang</footer>
<a class="entry-link" aria-label="post link to Learning with not Enough Data Part 3: Data Generation" href="https://jszzr.github.io/posts/2022-04-15-data-gen/"></a>
</article>
<article class="post-entry">
<header class="entry-header">
<h2>Learning with not Enough Data Part 2: Active Learning
</h2>
</header>
<section class="entry-content">
<p>This is part 2 of what to do when facing a limited amount of labeled data for supervised learning tasks. This time we will get some amount of human labeling work involved, but within a budget limit, and therefore we need to be smart when selecting which samples to label.
Notations Symbol Meaning $K$ Number of unique class labels. $(\mathbf{x}^l, y) \sim \mathcal{X}, y \in \{0, 1\}^K$ Labeled dataset....</p>
</section>
<footer class="entry-footer"><span title='2022-02-20 00:00:00 +0000 UTC'>February 20, 2022</span> · 22 min · Zhirui Zhang</footer>
<a class="entry-link" aria-label="post link to Learning with not Enough Data Part 2: Active Learning" href="https://jszzr.github.io/posts/2022-02-20-active-learning/"></a>
</article>
<article class="post-entry">
<header class="entry-header">
<h2>Learning with not Enough Data Part 1: Semi-Supervised Learning
</h2>
</header>
<section class="entry-content">
<p>When facing a limited amount of labeled data for supervised learning tasks, four approaches are commonly discussed.
Pre-training + fine-tuning: Pre-train a powerful task-agnostic model on a large unsupervised data corpus, e.g. pre-training LMs on free text, or pre-training vision models on unlabelled images via self-supervised learning, and then fine-tune it on the downstream task with a small set of labeled samples. Semi-supervised learning: Learn from the labelled and unlabeled samples together....</p>
</section>
<footer class="entry-footer"><span title='2021-12-05 00:00:00 +0000 UTC'>December 5, 2021</span> · 26 min · Zhirui Zhang</footer>
<a class="entry-link" aria-label="post link to Learning with not Enough Data Part 1: Semi-Supervised Learning" href="https://jszzr.github.io/posts/2021-12-05-semi-supervised/"></a>
</article>
<article class="post-entry">
<header class="entry-header">
<h2>How to Train Really Large Models on Many GPUs?
</h2>
</header>
<section class="entry-content">
<p>[Updated on 2022-03-13: add expert choice routing.] [Updated on 2022-06-10]: Greg and I wrote a shorted and upgraded version of this post, published on OpenAI Blog: “Techniques for Training Large Neural Networks”
In recent years, we are seeing better results on many NLP benchmark tasks with larger pre-trained language models. How to train large and deep neural networks is challenging, as it demands a large amount of GPU memory and a long horizon of training time....</p>
</section>
<footer class="entry-footer"><span title='2021-09-24 00:00:00 +0000 UTC'>September 24, 2021</span> · 21 min · Zhirui Zhang</footer>
<a class="entry-link" aria-label="post link to How to Train Really Large Models on Many GPUs?" href="https://jszzr.github.io/posts/2021-09-25-train-large/"></a>
</article>
<article class="post-entry">
<header class="entry-header">
<h2>What are Diffusion Models?
</h2>
</header>
<section class="entry-content">
<p>[Updated on 2021-09-19: Highly recommend this blog post on score-based generative modeling by Yang Song (author of several key papers in the references)]. [Updated on 2022-08-27: Added classifier-free guidance, GLIDE, unCLIP and Imagen. [Updated on 2022-08-31: Added latent diffusion model.
So far, I’ve written about three types of generative models, GAN, VAE, and Flow-based models. They have shown great success in generating high-quality samples, but each has some limitations of its own....</p>
</section>
<footer class="entry-footer"><span title='2021-07-11 00:00:00 +0000 UTC'>July 11, 2021</span> · 26 min · Zhirui Zhang</footer>
<a class="entry-link" aria-label="post link to What are Diffusion Models?" href="https://jszzr.github.io/posts/2021-07-11-diffusion-models/"></a>
</article>
<article class="post-entry">
<header class="entry-header">
<h2>Contrastive Representation Learning
</h2>
</header>
<section class="entry-content">
<p>The goal of contrastive representation learning is to learn such an embedding space in which similar sample pairs stay close to each other while dissimilar ones are far apart. Contrastive learning can be applied to both supervised and unsupervised settings. When working with unsupervised data, contrastive learning is one of the most powerful approaches in self-supervised learning.
Contrastive Training Objectives In early versions of loss functions for contrastive learning, only one positive and one negative sample are involved....</p>
</section>
<footer class="entry-footer"><span title='2021-05-31 00:00:00 +0000 UTC'>May 31, 2021</span> · 39 min · Zhirui Zhang</footer>
<a class="entry-link" aria-label="post link to Contrastive Representation Learning" href="https://jszzr.github.io/posts/2021-05-31-contrastive/"></a>
</article>
<article class="post-entry">
<header class="entry-header">
<h2>Reducing Toxicity in Language Models
</h2>
</header>
<section class="entry-content">
<p>Large pretrained language models are trained over a sizable collection of online data. They unavoidably acquire certain toxic behavior and biases from the Internet. Pretrained language models are very powerful and have shown great success in many NLP tasks. However, to safely deploy them for practical real-world applications demands a strong safety control over the model generation process.
Many challenges are associated with the effort to diminish various types of unsafe content:...</p>
</section>
<footer class="entry-footer"><span title='2021-03-21 00:00:00 +0000 UTC'>March 21, 2021</span> · 23 min · Zhirui Zhang</footer>
<a class="entry-link" aria-label="post link to Reducing Toxicity in Language Models" href="https://jszzr.github.io/posts/2021-03-21-lm-toxicity/"></a>
</article>
<article class="post-entry">
<header class="entry-header">
<h2>Controllable Neural Text Generation
</h2>
</header>
<section class="entry-content">
<p>[Updated on 2021-02-01: Updated to version 2.0 with several work added and many typos fixed.] [Updated on 2021-05-26: Add P-tuning and Prompt Tuning in the “prompt design” section.] [Updated on 2021-09-19: Add “unlikelihood training”.]
There is a gigantic amount of free text on the Web, several magnitude more than labelled benchmark datasets. The state-of-the-art language models (LM) are trained with unsupervised Web data in large scale. When generating samples from LM by iteratively sampling the next token, we do not have much control over attributes of the output text, such as the topic, the style, the sentiment, etc....</p>
</section>
<footer class="entry-footer"><span title='2021-01-02 00:00:00 +0000 UTC'>January 2, 2021</span> · 42 min · Zhirui Zhang</footer>
<a class="entry-link" aria-label="post link to Controllable Neural Text Generation" href="https://jszzr.github.io/posts/2021-01-02-controllable-text-generation/"></a>
</article>
<article class="post-entry">
<header class="entry-header">
<h2>How to Build an Open-Domain Question Answering System?
</h2>
</header>
<section class="entry-content">
<p>[Updated on 2020-11-12: add an example on closed-book factual QA using OpenAI API (beta).
A model that can answer any question with regard to factual knowledge can lead to many useful and practical applications, such as working as a chatbot or an AI assistant🤖. In this post, we will review several common approaches for building such an open-domain question answering system.
Disclaimers given so many papers in the wild:
Assume we have access to a powerful pretrained language model....</p>
</section>
<footer class="entry-footer"><span title='2020-10-29 00:00:00 +0000 UTC'>October 29, 2020</span> · 33 min · Zhirui Zhang</footer>
<a class="entry-link" aria-label="post link to How to Build an Open-Domain Question Answering System?" href="https://jszzr.github.io/posts/2020-10-29-odqa/"></a>
</article>
<article class="post-entry">
<header class="entry-header">
<h2>Neural Architecture Search
</h2>
</header>
<section class="entry-content">
<p>Although most popular and successful model architectures are designed by human experts, it doesn’t mean we have explored the entire network architecture space and settled down with the best option. We would have a better chance to find the optimal solution if we adopt a systematic and automatic way of learning high-performance model architectures.
Automatically learning and evolving network topologies is not a new idea (Stanley & Miikkulainen, 2002). In recent years, the pioneering work by Zoph & Le 2017 and Baker et al....</p>
</section>
<footer class="entry-footer"><span title='2020-08-06 00:00:00 +0000 UTC'>August 6, 2020</span> · 32 min · Zhirui Zhang</footer>
<a class="entry-link" aria-label="post link to Neural Architecture Search" href="https://jszzr.github.io/posts/2020-08-06-nas/"></a>
</article>
<article class="post-entry">
<header class="entry-header">
<h2>Exploration Strategies in Deep Reinforcement Learning
</h2>
</header>
<section class="entry-content">
<p>[Updated on 2020-06-17: Add “exploration via disagreement” in the “Forward Dynamics” section.
Exploitation versus exploration is a critical topic in Reinforcement Learning. We’d like the RL agent to find the best solution as fast as possible. However, in the meantime, committing to solutions too quickly without enough exploration sounds pretty bad, as it could lead to local minima or total failure. Modern RL algorithms that optimize for the best returns can achieve good exploitation quite efficiently, while exploration remains more like an open topic....</p>
</section>
<footer class="entry-footer"><span title='2020-06-07 00:00:00 +0000 UTC'>June 7, 2020</span> · 36 min · Zhirui Zhang</footer>
<a class="entry-link" aria-label="post link to Exploration Strategies in Deep Reinforcement Learning" href="https://jszzr.github.io/posts/2020-06-07-exploration-drl/"></a>
</article>
<article class="post-entry">
<header class="entry-header">
<h2>The Transformer Family
</h2>
</header>
<section class="entry-content">
<p>[Updated on 2023-01-27: After almost three years, I did a big refactoring update of this post to incorporate a bunch of new Transformer models since 2020. The enhanced version of this post is here: The Transformer Family Version 2.0. Please refer to that post on this topic.]
It has been almost two years since my last post on attention. Recent progress on new and enhanced versions of Transformer motivates me to write another post on this specific topic, focusing on how the vanilla Transformer can be improved for longer-term attention span, less memory and computation consumption, RL task solving and more....</p>
</section>
<footer class="entry-footer"><span title='2020-04-07 00:00:00 +0000 UTC'>April 7, 2020</span> · 25 min · Zhirui Zhang</footer>
<a class="entry-link" aria-label="post link to The Transformer Family" href="https://jszzr.github.io/posts/2020-04-07-the-transformer-family/"></a>
</article>
<article class="post-entry">
<header class="entry-header">
<h2>Curriculum for Reinforcement Learning
</h2>
</header>
<section class="entry-content">
<p>[Updated on 2020-02-03: mentioning PCG in the “Task-Specific Curriculum” section. [Updated on 2020-02-04: Add a new “curriculum through distillation” section.
It sounds like an impossible task if we want to teach integral or derivative to a 3-year-old who does not even know basic arithmetics. That’s why education is important, as it provides a systematic way to break down complex knowledge and a nice curriculum for teaching concepts from simple to hard....</p>
</section>
<footer class="entry-footer"><span title='2020-01-29 00:00:00 +0000 UTC'>January 29, 2020</span> · 24 min · Zhirui Zhang</footer>
<a class="entry-link" aria-label="post link to Curriculum for Reinforcement Learning" href="https://jszzr.github.io/posts/2020-01-29-curriculum-rl/"></a>
</article>
<article class="post-entry">
<header class="entry-header">
<h2>Self-Supervised Representation Learning
</h2>
</header>
<section class="entry-content">
<p>[Updated on 2020-01-09: add a new section on Contrastive Predictive Coding]. [Updated on 2020-04-13: add a “Momentum Contrast” section on MoCo, SimCLR and CURL.] [Updated on 2020-07-08: add a “Bisimulation” section on DeepMDP and DBC.] [Updated on 2020-09-12: add MoCo V2 and BYOL in the “Momentum Contrast” section.] [Updated on 2021-05-31: remove section on “Momentum Contrast” and add a pointer to a full post on “Contrastive Representation Learning”]...</p>
</section>
<footer class="entry-footer"><span title='2019-11-10 00:00:00 +0000 UTC'>November 10, 2019</span> · 38 min · Zhirui Zhang</footer>
<a class="entry-link" aria-label="post link to Self-Supervised Representation Learning" href="https://jszzr.github.io/posts/2019-11-10-self-supervised/"></a>
</article>
<footer class="page-footer">
<nav class="pagination">
<a class="next" href="https://jszzr.github.io/page/2/"> »</a>
</nav>
</footer>
</main>
<footer class="footer">
<span>© 2023 <a href="https://jszzr.github.io/">Lil'Log</a></span>
<span>
Powered by
<a href="https://gohugo.io/" rel="noopener noreferrer" target="_blank">Hugo</a> &
<a href="https://git.io/hugopapermod" rel="noopener" target="_blank">PaperMod</a>
</span>
</footer>
<a href="#top" aria-label="go to top" title="Go to Top (Alt + G)" class="top-link" id="top-link" accesskey="g">
<svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 12 6" fill="currentColor">
<path d="M12 6H0l6-6z" />
</svg>
</a>
<script>
let menu = document.getElementById('menu')
if (menu) {
menu.scrollLeft = localStorage.getItem("menu-scroll-position");
menu.onscroll = function () {
localStorage.setItem("menu-scroll-position", menu.scrollLeft);
}
}
document.querySelectorAll('a[href^="#"]').forEach(anchor => {
anchor.addEventListener("click", function (e) {
e.preventDefault();
var id = this.getAttribute("href").substr(1);
if (!window.matchMedia('(prefers-reduced-motion: reduce)').matches) {
document.querySelector(`[id='${decodeURIComponent(id)}']`).scrollIntoView({
behavior: "smooth"
});
} else {
document.querySelector(`[id='${decodeURIComponent(id)}']`).scrollIntoView();
}
if (id === "top") {
history.replaceState(null, null, " ");
} else {
history.pushState(null, null, `#${id}`);
}
});
});
</script>
<script>
var mybutton = document.getElementById("top-link");
window.onscroll = function () {
if (document.body.scrollTop > 800 || document.documentElement.scrollTop > 800) {
mybutton.style.visibility = "visible";
mybutton.style.opacity = "1";
} else {
mybutton.style.visibility = "hidden";
mybutton.style.opacity = "0";
}
};
</script>
<script>
document.getElementById("theme-toggle").addEventListener("click", () => {
if (document.body.className.includes("dark")) {
document.body.classList.remove('dark');
localStorage.setItem("pref-theme", 'light');
} else {
document.body.classList.add('dark');
localStorage.setItem("pref-theme", 'dark');
}
})
</script>
</body>
</html>