|
142 | 142 | <div id="pxlPagesContentParent" class="pxlPagesContentParentStyle"><div class="gpcpVisibleStyle procPagesContentStyle procPagesPlacementTripleStyle aiDevPageStyle pagesFader pagesVisOn"><div class="procPagesInnerBeforeBase procPagesInnerBefore"></div><div class="procPagesInnerStyle procPagesParentStyle aiDevPageParentStyle procPagesLayoutTripleStyle procPageNoMediaStyle" id="pxlPage_AIDev"><div class="procPageHeader procPagesHeaderStyle">AI Development</div><div class="procPagesHeaderLine aiDevPage-headerLine"></div><nav role="navigation" aria-label="Page sections" class="procPageSectionList aiDevPage-sectionNavListStyle"><div class="procPagesNavSectionStyle procPagesButtonStyle procPagesSectionNavColor aiDevPage-sectionNavButtonStyle" role="button" tabindex="0" aria-label="Navigate to My Introduction section">My Introduction</div><div class="procPagesNavSectionStyle procPagesButtonStyle procPagesSectionNavColor aiDevPage-sectionNavButtonStyle" role="button" tabindex="0" aria-label="Navigate to GAT & Language section">GAT & Language</div><div class="procPagesNavSectionStyle procPagesButtonStyle procPagesSectionNavColor aiDevPage-sectionNavButtonStyle" role="button" tabindex="0" aria-label="Navigate to ESN & Motion Prediction section">ESN & Motion Prediction</div><div class="procPagesNavSectionStyle procPagesButtonStyle procPagesSectionNavColor aiDevPage-sectionNavButtonStyle procPagesNavActive aiDevPage-sectionNavButtonActiveStyle" role="button" tabindex="0" aria-label="Navigate to Notes & Research section">Notes & Research</div></nav><section class="procPageMediaView aiDevPageScrollbarStyle" role="region" aria-label="Media gallery for AIDev page sections" aria-describedby="Dynamic media content that changes based on selected section" style="display: none;"><div class="procPagesMediaListStyle pagesVisOff"><iframe src="https://www.youtube-nocookie.com/embed/XJu-UJrI6yk" title="Useful AI for Visual Graphics" frameborder="0" allow="encrypted-media; picture-in-picture" referrerpolicy="strict-origin-when-cross-origin" allowfullscreen="true" class="procPagesMediaStyle procPagesLimitWidthStyle"></iframe></div></section><section class="procPageContentView aiDevPageScrollbarStyle" role="main" aria-label="Primary content area for AIDev page" aria-describedby="Main content that updates dynamically based on selected section navigation"><div class="procPageSectionContentStyle pagesVisOff" id="0"><div class="procPagesInnerContentStyle pagesVisOff"> |
143 | 143 | <div class="textSpacer"></div> |
144 | 144 | <div class="procPagesAboutMe-infoStyle"> |
145 | | - I started my dive into AI in 2008 writing a Boid / Crowd system for my thesis while in art college, School of Visual Arts. |
| 145 | + <br>I started my dive into AI in 2008 writing a Boid / Crowd system for my thesis while in art college, School of Visual Arts. |
146 | 146 | <br> It was an insane particle script + 3d animation cycles in Maya haha. |
147 | 147 | <br>Then I did Boid movement, navigation, & obstacle detection in animated films for 5 years at Blue Sky Studios, using Houdini. |
148 | 148 | <br> |
|
179 | 179 |
|
180 | 180 | <br><span class="innerCenter"> |
181 | 181 | You could guess, |
182 | | - <br>All the recent AI stuff has been quite serendipitous for my creation! |
| 182 | + <br>All the recent AI hullabaloo (2019-...) |
| 183 | + <br>Has been quite serendipitous for my creation! |
183 | 184 | </span> |
184 | 185 | </div> |
185 | 186 | </div></div><div class="procPageSectionContentStyle pagesVisOn procPagesSectionActive" id="3"><div class="procPagesInnerContentStyle procPagesSectionActive pagesVisOn"> |
186 | 187 | <div class="textSpacer"></div> |
187 | 188 | <div class="procPagesAboutMe-infoStyle"> |
188 | | - <br>If you couldn't tell, I'm training my AIs on my own works. |
| 189 | + I made a whole 'blog' system for this... yet here we are.... |
| 190 | + <br> ( nudge, nudge ) |
| 191 | + |
| 192 | + <br><br><div class="procPagesAboutMeBar"></div> |
| 193 | + |
| 194 | + <br><br>If you couldn't tell, I'm training my AIs on my own works. |
189 | 195 | <br>A personally made AI trained on personally made images / videos / photos / code / writing. |
190 | 196 | <br> That means I can copyright my generations, right? |
191 | 197 | <br> If I made every aspect of the AI & training data? |
|
195 | 201 |
|
196 | 202 | <br><div class="procPagesAboutMeBar"></div> |
197 | 203 |
|
198 | | - <br>I've begun on the core of the AI, as of May 24th, 2025. |
| 204 | + <br><br>I've begun on the core of the AI, as of May 24th, 2025. |
199 | 205 | <br> I have the beginnings of a 'Micro-Term' memory implemented to act as a gated-attention during inference. |
200 | 206 | <br>This, paired with automatic graph edge splitting ('Dynamic' in DGNN or DGAT) and use of geometric clustering, seems to be giving me values of a "remembered" object when it's outside of the dataset. |
201 | | - <br> Bodily awarness of limbs, objects outside of the field of view, and other 'long term' tensors/classifications at a temporary scale. |
| 207 | + <br> Hopefully leading to bodily awareness of limbs, objects outside of the field of view, and other 'long term' tensors/classifications at a temporary scale. |
202 | 208 | <br> |
203 | 209 | <br>It's a 4d kernel, in that it uses an ESN to train on it's own mistakes, |
204 | 210 | <br> Basing it's decisions on prior back-propagation states/adjustments. |
205 | | - <br> The beginnings of a meta-learning process, hehe. |
| 211 | + <br> The beginnings of a meta-learning process, I hope! |
206 | 212 | <br> |
207 | 213 | <br>I'm using a method I'm calling 'Stack Crunching', |
208 | 214 | <br> Where I agregate the time dependent weights into a "checkpoint" of sorts. |
|
228 | 234 | <br> I am a Technical Artist first before an AI Researcher after all, BOIDS! |
229 | 235 | <br> |
230 | 236 | <br>It all comes down to BOOOOIIIIIDDDDSSSSSS instead of Adam, baby! |
231 | | - <br>Tiny Brains running around in hyperdimensional space like little buggers running around avoiding each other. |
| 237 | + <br>Because, what is Adam? It's a direction to move in a field of numbers, with momentum and a learning rate. |
| 238 | + <br> Yet... That's just a simple Boid, now isn't it? |
| 239 | + <br> Just without a few of the more advanced rules, which make boids feel so alive! |
| 240 | + <br> |
| 241 | + <br>Having some Tiny Brains running around in hyperdimensional space like little buggers running around avoiding each other. |
232 | 242 | <br> Because if they collide, double activation happens when it may not be desired. |
233 | 243 | <br> (I'm happy I finally saw a paper on Tiny Brains, giving my idea credence, cause it fits! .. in my mind.) |
234 | 244 | <br>(Only difference is that it was a <a href="https://www.nature.com/articles/s41586-025-09142-4" target="_blank">study</a> into small biological systems, not artificial ones... but I'm gettin there!) |
235 | 245 | <div class="textFullRight">- July 2025</div> |
236 | 246 |
|
237 | 247 | <br><br><div class="procPagesAboutMeBar"></div> |
| 248 | + |
| 249 | + <br>I'd like to believe I'm moving in the right direction with the feedback systems I'm developing. |
| 250 | + <br> But been further creating other architectures to see how they operate. |
| 251 | + |
| 252 | + <br><br>I created a GAN for upressing, which helped me understand a bit better the pairing of mental structures between both our brain's hemispheres. |
| 253 | + <br> So I added a time based memory to check if the training was moving in the right direction. |
| 254 | + <br> It definitely helped guide training a bit quicker. |
| 255 | + |
| 256 | + <br><br>Shows my knowledge base that I'm impressed by back-up supported learning... |
| 257 | + <br> But'is proof of concept! |
238 | 258 |
|
239 | | - <br> |
240 | | - <br> With a bit more research into the types of minds that brought us DeepMind, and their work on GNN networks, |
241 | | - <br> I read a bit of Petar Velickovic's work on topological deep learning and the geometry of GNNs. |
242 | | - <br> Coming to find out that my idea of 'Stack Crunching' is similar to 'Squashing' in GNNs. |
243 | | - <br> |
244 | | - <br> So I've been inspired to propperly name my neural structure- |
245 | | - <br>It's a <span class="textName">Dynamic Pointer-Attention Message Passing Neural Network with Affine-Projections</span> |
246 | | - <br> or a <span class="textName">dPA-MPNN</span> |
247 | | - |
248 | | - <br><br> But I must say, this isn't Affine Projections like in the papers, |
249 | | - <br> It's more like a 'projection' of the data into 'pointer' space; |
250 | | - <br> Actual Affine Matricies. |
251 | | - <br> I am a Technical Artist first before an AI Researcher after all, BOIDS! |
252 | | - <br> |
253 | | - <br>It all comes down to BOOOOIIIIIDDDDSSSSSS instead of Adam, baby! |
254 | | - <br>Tiny Brains running around in hyperdimensional space like little buggers running around avoiding each other. |
255 | | - <br> Because if they collide, double activation happens when it may not be desired. |
256 | | - <br> (I'm happy I finally saw a paper on Tiny Brains, giving my idea credence, cause it fits! .. in my mind.) |
257 | | - <br>(Only difference is that it was a <a href="https://www.nature.com/articles/s41586-025-09142-4" target="_blank">study</a> into small biological systems, not artificial ones... but I'm gettin there!) |
| 259 | + <br><br>Adversarial networks exist in nature to guide a 'single' thought's path. |
| 260 | + <br> Yet in the case of Group Think between humans, |
| 261 | + <br> Balance is never reached. |
258 | 262 | <div class="textFullRight">- August 2025</div> |
259 | 263 | <br> |
260 | 264 | </div> |
|
0 commit comments