Mapping the Drosophila Eye

 To emulate the Drosophila eye sensory system, I had to map what Ommatidia to the optical sensory neurons. I used the Flywire.ai right eye map from this map: Retina Grid. Taking each ommatidia, I first had to map the column id associated to the ommatidia. Here is that map by column id:

I was able to create the eye map into a grid that has 28 columns and 28 rows, and put this grid into a MS SQL table:

C1 C2 C3 C4 C5 C6 C7 C8 C9 C10 C11 C12 C13 C14 C15 C16 C17 C18 C19 C20 C21 C22 C23 C24 C25 C26 C27 C28
0 0 0 0 0 0 0 0 787 0 660 207 494 448 476 380 65 295 352 729 41 784 783 0 0 0 0 0
0 0 0 0 0 0 796 788 402 438 322 767 631 138 282 202 186 237 223 343 222 645 8 782 781 0 0 0
0 0 0 0 779 778 629 453 324 570 353 472 621 284 329 680 618 707 625 693 753 744 692 442 718 780 0 0
0 0 0 0 775 584 587 252 105 359 310 440 172 336 356 582 393 428 567 493 398 473 518 702 303 273 0 0
0 0 777 776 756 414 451 187 425 257 78 147 658 456 39 547 404 50 160 370 133 70 748 720 622 412 131 0
0 0 34 226 94 755 679 388 33 216 586 668 607 737 166 711 462 21 689 260 378 639 292 360 43 386 644 675
0 646 30 463 253 357 165 503 68 302 326 482 601 760 31 168 5 457 696 358 553 340 143 188 301 684 713 19
759 217 643 700 53 99 573 605 348 66 523 350 466 127 762 476 42 418 429 616 661 246 541 46 763 539 75 311
728 189 415 721 714 80 63 102 58 374 130 623 772 745 32 261 421 176 84 738 437 232 299 287 200 1 309 469
192 339 498 545 499 235 715 126 405 178 218 87 608 677 556 289 396 510 479 468 489 742 509 508 278 383 695 106
496 575 36 125 502 9 630 581 140 230 335 590 69 92 663 599 196 504 445 765 654 555 279 122 308 731 210 12
727 215 569 561 28 365 271 306 546 528 683 648 751 377 636 638 91 752 119 434 732 392 242 432 490 375 239 82
88 304 274 670 770 495 364 768 536 673 132 37 588 224 145 649 236 319 115 206 263 90 283 298 193 574 443 227
291 517 198 327 229 694 724 330 534 276 73 234 501 136 747 22 447 550 59 746 477 54 115 214 97 93 267 221
730 497 512 325 458 597 464 118 465 79 734 705 531 486 628 250 338 749 609 703 426 538 522 170 699 112 248 228
665 655 81 758 204 552 602 203 272 662 529 487 372 627 38 120 164 177 410 195 424 475 162 45 562 265 384 211
363 420 219 459 316 391 49 591 213 15 766 26 181 409 701 29 543 71 6 485 76 240 61 245 317 687 408 690
268 371 635 369 205 109 716 158 382 77 220 17 146 194 757 589 264 320 40 190 315 651 659 519 337 184 7 259
500 698 682 137 566 312 481 678 416 483 740 124 270 361 433 157 116 413 614 430 209 647 527 86 626 175 568 615
182 521 557 328 44 470 444 484 600 450 100 452 96 549 526 74 24 540 431 439 579 551 514 685 407 390 417 346
595 57 3 511 454 578 231 20 345 185 249 262 506 656 199 191 98 723 403 419 108 55 244 642 507 652 653 148
11 542 149 563 577 233 161 544 593 114 446 56 139 129 571 754 251 67 741 530 151 513 674 347 349 255 280 323
594 722 449 10 18 174 669 152 441 342 212 331 548 436 533 387 640 344 16 634 23 247 596 183 179 367 0 0
516 341 285 435 427 293 300 379 423 241 598 624 296 167 169 351 611 617 142 474 385 258 761 733 134 604 0 0
0 743 171 773 256 218 524 480 632 269 362 89 294 51 197 606 525 401 467 397 460 619 368 411 366 0 0 0
0 795 505 208 686 95 85 637 128 83 717 373 697 333 395 243 52 681 520 676 135 558 48 4 0 0 0 0
0 0 0 141 290 64 750 564 254 610 111 735 35 537 286 585 633 580 709 603 725 671 399 0 0 0 0 0
0 0 0 307 794 710 332 583 471 173 225 406 334 27 565 266 13 657 354 650 612 708 0 0 0 0 0 0

where "0" (Zero) is a null ommatidia, or an ommatidia that does not exist.

Using the "Visual Neuron Column" data from Flywire, I was able to load into an MS SQL database and then I could map the column ids (ommatidia) to the sensory neurons. As an example, using the first ommatidia id in column 1 (C1) is 759. We can do a join to find the sensory neurons that are part of this id:


Using the pixels in a 28x28 image, I then used the pixel color to determine is that ommatidia should be stimulated. We have a set of 28x28 images in the MNIST dataset:












Therefore, where the image has a white pixel, we find the column id from the Eyemap database table for that pixel in the grid, and then we find the neurons that are associated and stimulate those neurons. 

In this image, we see the first white pixel is located at column 11, row 7 = column id 326:

Using this SQL Query from the Visual Neuron Columns data:

We find the neurons that are associated to that column in the eye map:



I created a program that stimulates these neurons using the MNIST dataset where the pixels are white and stimulates several times as if the fly was actually looking at the image and I capture each stimulation of the entire Drosophila central nervous system.

From this, I am trying to determine if the Drosophila CNS can discriminate these images and I am using the rest of the MNIST dataset to find out if invariance exists using this technique. The raster>stimulated captured data is quite huge so it is taking time to analyze all of the captured data.

For invariance, here is an example of the augmented images that I am having the Drosophila CNS emulation review:








If the Drosophila CNS emulation can detect the images above as Zeros, this will show there is mechanisms in the CNS that can discriminate these images as Zeros as opposed to other images representing other numbers. 

The next big question will be what are those mechanisms?

Comments

Post a Comment

Popular posts from this blog

Use Cases for the emulation of the Drosophila nervous system

 I am often asked, especially by VC types, what can you do with animal emulations? What they are really asking is how can this be applied and how can we make money? I will admit, emulating a nervous system is like telling people you are developing a new computer operating system. There are thousands of things you can run on an operating system so asking for the One thing, is analogous to asking what is the one thing you can run on your new OS that stands apart from all others? There is no One thing but thousands of things.  If I emulated a complete human level system, what is the One thing it can do that would exceed all others? The human brain is capable of thousands and thousands of things it can do. I think you get the picture.  The One thing that I have seen with all animal nervous system emulations is generalized intelligence. The one thing that current AI cannot accomplish which is being able to quickly adapt to any given situation and as we climb the evolutionary ...
Read more

Making Dronesophila Part 1

Image
       E mulating the Drosophila nervous system was step one which I did successfully on a single laptop using functional groups. This works VERY well.   The functional groups I created were: Neuron Group    Synapse Count central                          1455116 optic_lobes                1217922 visual_projection      268547 AN                                    115292 Kenyon_Cell                92875 ALPN                          88302 CX                                    85831...
Read more

How do we use the Connectome for AGI?

Image
Using the connectome for AGI is the tricky part of the process. Tricky because it does take some thought behind how to associate sensory neurons to real world concepts and how to read the motor output in order to determine the results. To do this we break down the process simply as Sensory Input > Cortical Processing > Motor Output ( > Sensory Processing > Cortical Processing > ...). Sensory Processing Each connectome has a set of sensory neurons. When we create our AGI, we connect weights to the sensory neurons based on how we perceive the sensory input on a particular sensory neuron. As an example, if we want our AGI system to avoid particular situations, we connect that sensory input to sensory neurons that would cause the connectome to sense pain or unpleasant conditions. If we want our AGI system to want or like a particular input, we connect that sensory input to sensory neurons that would cause pleasant conditions in the connectome. Using Robotics makes this p...
Read more