Repository navigation
Expand file tree
/
Copy pathresearch.html
More file actions
530 lines (518 loc) · 19.7 KB
/
Copy pathresearch.html
File metadata and controls
530 lines (518 loc) · 19.7 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
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
<!doctype html>
<html lang="en">
<head>
<meta charset="UTF-8" />
<meta name="viewport" content="width=device-width, initial-scale=1.0" />
<meta name="theme-color" content="#fafaf7" />
<meta
name="description"
content="Predictive Sparse Intelligence: TPH to MR-TPH to Spectrum-State to Imugi. A research program on systems that predict what information, computation, memory, precision, and hardware will be necessary."
/>
<link rel="canonical" href="https://nitg3n.xyz/research.html" />
<meta property="og:title" content="Predictive Sparse Intelligence" />
<meta
property="og:description"
content="Predictive Sparse Intelligence: TPH to MR-TPH to Spectrum-State to Imugi. A research program on systems that predict what information, computation, memory, precision, and hardware will be necessary."
/>
<meta property="og:type" content="website" />
<meta property="og:url" content="https://nitg3n.xyz/research.html" />
<meta name="twitter:card" content="summary" />
<link rel="icon" href="favicon.svg" type="image/svg+xml" />
<link rel="stylesheet" href="styles.css" />
<title>Predictive Sparse Intelligence | Jay Han</title>
</head>
<body>
<header class="top">
<div class="top-inner">
<a class="wordmark" href="index.html">Jay Han</a>
<nav class="top-nav" aria-label="Site">
<a class="top-link" href="research.html">Research</a>
</nav>
</div>
</header>
<main class="wrap">
<header class="hero">
<h1>Predictive Sparse Intelligence</h1>
<p class="subtitle">
A long-running research program exploring a simple question:
</p>
</header>
<blockquote class="quote">
<p>
Can intelligence avoid storing every piece of information, processing
every token at the same depth, and activating every parameter all
the time?
</p>
</blockquote>
<section aria-label="Lineage">
<div class="block">
<p>The project evolved through four generations:</p>
<div class="gen-grid">
<div class="gen">
<p class="gen-name">TPH</p>
<p class="gen-s">State Sparsity</p>
</div>
<div class="arrow" aria-hidden="true">→</div>
<div class="gen">
<p class="gen-name">MR-TPH</p>
<p class="gen-s">Temporal Sparsity</p>
</div>
<div class="arrow" aria-hidden="true">→</div>
<div class="gen">
<p class="gen-name">Spectrum-State</p>
<p class="gen-s">Computational Sparsity</p>
</div>
<div class="arrow" aria-hidden="true">→</div>
<div class="gen">
<p class="gen-name">Imugi</p>
<p class="gen-s">Operational Sparsity</p>
</div>
</div>
<p class="scope">
State → Time → Compute → Memory / Capacity / Precision / Hardware
</p>
</div>
<div class="block">
<p>The central idea has remained the same:</p>
<blockquote class="quote">
<p>
<strong>Intelligence is not the ability to process everything.</strong><br />
<strong>It is the ability to predict what does not need to be processed.</strong>
</p>
</blockquote>
</div>
</section>
<section>
<h2 class="section-head"><span>TPH</span><span>01</span></h2>
<div class="block">
<p>
<strong>TPH (Top-down Predictive Hierarchy)</strong> began with
predictive coding.
</p>
<p>
Instead of forwarding all information upward, higher layers predict
lower-layer states and only unexplained information — prediction
error — is propagated.
</p>
<p>
The project later found that filtering error alone was not enough.
The more important problem was what the model stored in its
internal state.
</p>
<p>This led to <strong>State Sparsity</strong>:</p>
<ul class="body-list">
<li>Predict first</li>
<li>Update from error</li>
<li>Store only important information</li>
<li>Keep higher-level representations sparse</li>
<li>Treat prediction as an information bottleneck</li>
</ul>
</div>
<div class="block">
<div class="flow">
<div class="flow-box">
<p class="eyebrow">Higher layer</p>
<p>Predicts lower-layer states</p>
</div>
<div class="arrow" aria-hidden="true">→</div>
<div class="flow-box">
<p class="eyebrow">Lower layer</p>
<p>Propagates only prediction error</p>
</div>
</div>
<p class="note">Predict first. Propagate only the error.</p>
</div>
<div class="block">
<p>The core principle was:</p>
<blockquote class="quote">
<p><strong>If something is already predictable, do not process it again.</strong></p>
</blockquote>
</div>
</section>
<section>
<h2 class="section-head"><span>MR-TPH</span><span>02</span></h2>
<div class="block">
<p>
TPH reduced unnecessary state updates, but its recurrent structure
remained difficult to parallelize efficiently on GPUs.
</p>
<p>
<strong>MR-TPH (Multi-Resolution TPH)</strong> extended sparsity
from state to time.
</p>
<p>Instead of processing an entire sequence sequentially at full resolution:</p>
<ul class="body-list">
<li>Fine-grained information is processed in parallel chunks</li>
<li>Each chunk produces a summary and boundary state</li>
<li>Only the lower-resolution summary sequence is processed sequentially</li>
<li>Prediction error still controls state updates</li>
</ul>
</div>
<div class="block">
<div class="flow flow--stack">
<div class="flow-box">
<p class="eyebrow">Fine-grained · parallel</p>
<p>Each chunk produces a summary and boundary state</p>
</div>
<div class="arrow" aria-hidden="true">↓</div>
<div class="flow-box">
<p class="eyebrow">Coarse-grained · sequential</p>
<p>Only the summary sequence is processed sequentially</p>
</div>
</div>
</div>
<div class="block">
<p>This created a new form of sparsity:</p>
<blockquote class="quote">
<p><strong>Not every moment needs the same temporal resolution.</strong></p>
</blockquote>
</div>
<div class="block">
<p>
An MR-TPH implementation, <strong>Raptor</strong>, was evaluated on
WikiText-2 under a specific experimental setup.
</p>
<div class="table-wrap">
<table class="table">
<thead>
<tr>
<th scope="col">Model</th>
<th scope="col" class="num">Test PPL</th>
<th scope="col" class="num">Epoch Time</th>
</tr>
</thead>
<tbody>
<tr>
<td>Transformer</td>
<td class="num">612.45</td>
<td class="num">~13.8s</td>
</tr>
<tr>
<td>Raptor / MR-TPH</td>
<td class="num">355.90</td>
<td class="num">~15.0s</td>
</tr>
</tbody>
</table>
</div>
<p class="note">
These results belong to that specific comparison and should not be
interpreted as a general claim of superiority over Transformers.
</p>
</div>
</section>
<section>
<h2 class="section-head"><span>Spectrum-State</span><span>03</span></h2>
<div class="block">
<p>
MR-TPH separated fine-grained and coarse-grained time processing.
</p>
<p>
<strong>Spectrum-State</strong> extended the same philosophy to
functions.
</p>
<p>
Instead of asking one mechanism to handle global context,
sequential state, exact recall, and reasoning equally well, the
architecture separated those roles.
</p>
</div>
<div class="block">
<h3>Early Spectrum-State</h3>
<p>The architecture evolved around several components:</p>
<div class="roles">
<div class="role">
<p class="eyebrow">Eye</p>
<p>Global or spectral context</p>
</div>
<div class="role">
<p class="eyebrow">Hand</p>
<p>Mamba / SSM-based sequence processing</p>
</div>
<div class="role">
<p class="eyebrow">Brain</p>
<p>Global context control</p>
</div>
<div class="role">
<p class="eyebrow">Shield</p>
<p>Selective attention for exact recall</p>
</div>
</div>
<p>A key development was <strong>GCS Sleep / Wake</strong>:</p>
<div class="flow">
<div class="flow-box">
<p class="eyebrow">Sleep</p>
<p>Reuse existing global context when it is sufficient</p>
</div>
<div class="arrow" aria-hidden="true">⇄</div>
<div class="flow-box">
<p class="eyebrow">Wake</p>
<p>Recompute global perception only when uncertainty rises</p>
</div>
</div>
<p>This extended the original idea from:</p>
<p class="scope">Sparse State → Sparse Compute</p>
</div>
<div class="block">
<h3>Spectrum-State v9</h3>
<p>
Spectrum-State v9 accepted that SSMs can lose exact long-range
details when compressing history into fixed-size state.
</p>
<p>
Instead of removing attention entirely, it used attention
selectively.
</p>
<p>The reported configuration included:</p>
<ul class="body-list">
<li>85M parameters</li>
<li>12 layers</li>
<li>9 Mamba layers</li>
<li>3 selective attention layers</li>
<li>32k evaluation context</li>
</ul>
<div
class="layers"
role="img"
aria-label="12 layers: 9 Mamba layers and 3 selective attention layers"
>
<span class="l-m"></span><span class="l-m"></span><span class="l-m"></span><span class="l-m"></span><span class="l-m"></span><span class="l-m"></span><span class="l-m"></span><span class="l-m"></span><span class="l-m"></span><span class="l-a"></span><span class="l-a"></span><span class="l-a"></span>
</div>
<p class="note">
9 Mamba layers carry sequential state; 3 selective attention
layers handle exact recall.
</p>
<p>In a reported 32k Needle-in-a-Haystack test:</p>
<div class="table-wrap">
<table class="table">
<thead>
<tr>
<th scope="col">Model</th>
<th scope="col" class="num">Success Rate</th>
</tr>
</thead>
<tbody>
<tr>
<td>Mamba-2</td>
<td class="num">83.3%</td>
</tr>
<tr>
<td>Spectrum-State v9</td>
<td class="num">100%</td>
</tr>
</tbody>
</table>
</div>
<p class="note">
The test used <strong>N=30</strong> with the needle approximately
<strong>29k tokens away</strong>.
</p>
</div>
<div class="block">
<h3>Spectrum-State v14.5–v15</h3>
<p>Later versions expanded selective computation further:</p>
<ul class="body-list">
<li>Adaptive perception</li>
<li>Sparse experts</li>
<li>Compressed memory</li>
<li>Dynamic reasoning depth</li>
<li>Critic / PRM-style verification</li>
<li>Fast and slow reasoning paths</li>
</ul>
<p>The question changed from:</p>
<blockquote class="quote">
<p>“When should the model update its state?”</p>
</blockquote>
<p>to:</p>
<blockquote class="quote">
<p><strong>“When should the model think harder?”</strong></p>
</blockquote>
</div>
</section>
<section>
<h2 class="section-head"><span>Imugi</span><span>04</span></h2>
<div class="block">
<p>
<strong>Imugi</strong> extends the same research lineage from
architecture into system-level resource allocation.
</p>
<p>
The goal is no longer only to react when a difficult input appears.
</p>
<p>
The system attempts to predict what resources will be needed next.
</p>
<p>This includes:</p>
<ul class="body-list">
<li>Multi-timescale recurrent processing</li>
<li>Compressed associative memory</li>
<li>Persistent semantic memory</li>
<li>Generalist / Specialist routing</li>
<li>Critic-based foresight</li>
<li>Hybrid precision</li>
<li>Specialist prefetching</li>
<li>Hardware-aware execution</li>
<li>Long-context state carry</li>
<li>Distributed training</li>
</ul>
</div>
<div class="block">
<p>The key transition is:</p>
<p class="scope">Reactive Compute → Predictive Compute</p>
<p>
Instead of only allocating more computation after difficulty
appears, the system aims to anticipate future difficulty and
prepare:
</p>
<div class="flow">
<div class="flow-box">
<p class="eyebrow">Prepare</p>
<p>The right resources for future difficulty — even hardware.</p>
</div>
<div class="arrow" aria-hidden="true">→</div>
<div class="chips">
<span>Specialist</span><span>Memory</span><span>Precision</span><span>Compute path</span><span>Hardware</span>
</div>
</div>
</div>
</section>
<section>
<h2 class="section-head"><span>The Research Lineage</span><span>05</span></h2>
<div class="block">
<div class="table-wrap">
<table class="table">
<thead>
<tr>
<th scope="col">Generation</th>
<th scope="col">Main Question</th>
<th scope="col">Form of Sparsity</th>
</tr>
</thead>
<tbody>
<tr>
<td>TPH</td>
<td>What should be stored?</td>
<td>Representation / State Sparsity</td>
</tr>
<tr>
<td>MR-TPH</td>
<td>What should be processed sequentially?</td>
<td>Temporal Sparsity</td>
</tr>
<tr>
<td>Spectrum-State</td>
<td>Which mechanisms and how much compute should be used?</td>
<td>Computational Sparsity</td>
</tr>
<tr>
<td>Imugi</td>
<td>What memory, capacity, precision, and hardware should be prepared?</td>
<td>Operational Sparsity</td>
</tr>
</tbody>
</table>
</div>
</div>
<div class="block">
<p>The broader conceptual evolution is:</p>
<p class="scope">
Prediction → Error → Sparse State → Multi-Resolution → Global
Context → Exact Recall → Adaptive Compute → Sparse Capacity →
Persistent Memory → Foresight → Hardware-Aware Intelligence
</p>
</div>
</section>
<section>
<h2 class="section-head"><span>Experimental Work</span><span>06</span></h2>
<div class="block">
<p>
Parts of the research program have been implemented and evaluated.
</p>
</div>
<div class="block">
<h3>Spectrum-State v7</h3>
<p>Reported training experiment:</p>
<ul class="body-list">
<li>TinyStories</li>
<li>50,000 steps</li>
<li>Loss: 0.54</li>
<li>VRAM usage: 1.8 GB</li>
</ul>
</div>
<div class="block">
<h3>Raptor / MR-TPH</h3>
<p>Reported WikiText-2 comparison:</p>
<ul class="body-list">
<li>Raptor Test PPL: 355.90</li>
<li>Transformer Test PPL: 612.45</li>
</ul>
</div>
<div class="block">
<h3>Spectrum-State v9</h3>
<p>Reported long-context recall experiment:</p>
<ul class="body-list">
<li>85M parameters</li>
<li>32k context</li>
<li>N=30</li>
<li>~29k token retrieval distance</li>
<li>Spectrum-State: 100%</li>
<li>Mamba-2: 83.3%</li>
</ul>
<p class="note">
These results are specific to their respective experimental setups.
</p>
</div>
</section>
<section>
<h2 class="section-head"><span>Current Research Status</span><span>07</span></h2>
<div class="block">
<p>
The earlier generations include implemented experiments and
quantitative evaluations.
</p>
<p>
Imugi currently represents a broader architectural and systems
specification, including:
</p>
<ul class="body-list">
<li>memory policy,</li>
<li>specialist routing,</li>
<li>foresight control,</li>
<li>hybrid precision,</li>
<li>training stability,</li>
<li>distributed execution,</li>
<li>and hardware orchestration.</li>
</ul>
</div>
<div class="block">
<p>Several components still require dedicated benchmarking, including:</p>
<ul class="body-list">
<li>long-context memory retention,</li>
<li>specialist routing efficiency,</li>
<li>foresight accuracy,</li>
<li>hardware prefetch gains,</li>
<li>low-precision stability,</li>
<li>and end-to-end Imugi performance.</li>
</ul>
</div>
</section>
<section>
<h2 class="section-head"><span>Summary</span><span>08</span></h2>
<div class="block">
<blockquote class="quote">
<p>
<strong>Predictive Sparse Intelligence is a research program about
building systems that predict not only what comes next, but what
information, computation, memory, precision, and hardware
resources will actually be necessary to handle it.</strong>
</p>
</blockquote>
</div>
</section>
</main>
<footer class="site-foot">
<span>nitg3n.xyz</span>
<span>Project Default · Pre-seed</span>
</footer>
</body>
</html>