SlideShare une entreprise Scribd logo
1  sur  25
Télécharger pour lire hors ligne
Sequential Processing in Nature,
‘Anything but’ in Scientific Computation
an observation
Dominic Jones
Netherhall House, London
January 2018
Nature - ‘It should just work’
Nature - ‘It should just work’
Dominic Jones (Netherhall House, London) Sequential Processing in Nature,‘Anything but’ in Scientific ComputationJanuary 2018 2 / 25
Nature - ‘It should just work’
Protein folding
Synthesising the chain
Finding the global minimum
Folding occurs in microseconds, but cannot be predicted
Dominic Jones (Netherhall House, London) Sequential Processing in Nature,‘Anything but’ in Scientific ComputationJanuary 2018 3 / 25
Nature - ‘It should just work’
Abnormal folding
Deformed proteins cannot be mended
Dominic Jones (Netherhall House, London) Sequential Processing in Nature,‘Anything but’ in Scientific ComputationJanuary 2018 4 / 25
Nature - ‘It should just work’
In brief
1 From a chain of amino acids a specific highly complex shape is
required
2 There is no apparent error correction in the process (or very little)
3 It is usually successful
4 When it isn’t, Creutzfeldt-Jakob disease, Parkinson’s, Alzheimer’s,
etc
Dominic Jones (Netherhall House, London) Sequential Processing in Nature,‘Anything but’ in Scientific ComputationJanuary 2018 5 / 25
Scientific computation - ‘T = abs(T)’
Scientific computation - ‘T = abs(T)’
Dominic Jones (Netherhall House, London) Sequential Processing in Nature,‘Anything but’ in Scientific ComputationJanuary 2018 6 / 25
Scientific computation - ‘T = abs(T)’
Attempts at prediction and projection
Numerical approximation Local sensitivities
Iterative, discretised and linearised algorithms
Dominic Jones (Netherhall House, London) Sequential Processing in Nature,‘Anything but’ in Scientific ComputationJanuary 2018 7 / 25
Scientific computation - ‘T = abs(T)’
All starts with a graph (mesh)
Resolve change at the small scales
Mesh implies graph, implies matrix, implies matrix inversion
Dominic Jones (Netherhall House, London) Sequential Processing in Nature,‘Anything but’ in Scientific ComputationJanuary 2018 8 / 25
Scientific computation - ‘T = abs(T)’
Matrix inversion: scaling up
void inverse(int n, float a[][2*n_max])
{
for (int i = 0; i < n; i++)
for (int j = n; j < 2*n; j++)
a[i][j] = (i == j-n? 1: 0);
for (int i = 0; i < n; i++) { // linear
float aii = a[i][i];
for (int j = i; j < 2*n; j++)
a[i][j] = a[i][j] / aii;
for (int j = 0; j < n; j++) { // quadratic!
if (i != j) {
float aji = a[j][i];
for (int k = 0; k < 2*n; k++) // cubic!!
a[j][k] = a[j][k] - aji * a[i][k]; }}}}
Dominic Jones (Netherhall House, London) Sequential Processing in Nature,‘Anything but’ in Scientific ComputationJanuary 2018 9 / 25
Scientific computation - ‘T = abs(T)’
Computation: slow but compact
Geometry and mesh Connectivity graph
Dominic Jones (Netherhall House, London) Sequential Processing in Nature,‘Anything but’ in Scientific ComputationJanuary 2018 10 / 25
Scientific computation - ‘T = abs(T)’
Topology Representation
Symmetric, sparse (48 non-zeros), irregular


a0,0 a0,1 a0,8
a1,0 a1,1 a1,2 a1,6
a2,1 a2,2 a2,3
a3,2 a3,3 a3,4
a4,3 a4,4 a4,5 a4,13
a5,4 a5,5 a5,6 a5,11
a6,1 a6,5 a6,6 a6,7
a7,6 a7,7 a7,8 a7,9
a8,0 a8,7 a8,8
a9,7 a9,9 a9,10
a10,9 a10,10 a10,11
a11,5 a11,10 a11,11 a11,12
a12,11 a12,12 a12,13
a13,4 a13,12 a13,13


Dominic Jones (Netherhall House, London) Sequential Processing in Nature,‘Anything but’ in Scientific ComputationJanuary 2018 11 / 25
Scientific computation - ‘T = abs(T)’
Sparsity and Indirection
Dense storage: 196 values, 24% efficient
Direct access to values
float[14][14] A;
A[2][3] = 3.142;
Compressed row storage: 111 values, 56% efficient
Requires indirection to access values (10x slower)
int[15] IA = [0, 3, 7, 10, ...];
int[48] JA = [0, 1, 8, 0, 1, 2, 6, 1, 2, 3, ...];
float[48] A;
A[JA[IA[2]+2]] = 3.142; // i.e. A[2][3] = 3.142;
Dominic Jones (Netherhall House, London) Sequential Processing in Nature,‘Anything but’ in Scientific ComputationJanuary 2018 12 / 25
Scientific computation - ‘T = abs(T)’
In brief
1 Many scientific algorithms have graphs and matrices, and perform
matrix inversion at their core
2 They are often woefully inefficient due to the requirement to use
compressed storage of irregular data
3 The compromise on efficiency is to leverage compact memory
footprint
4 In the process of linearising, much of the ‘beauty’ of underlying the
mathematics gives way to crude approximation
Dominic Jones (Netherhall House, London) Sequential Processing in Nature,‘Anything but’ in Scientific ComputationJanuary 2018 13 / 25
Bridging the Nature — Computation dichotomy
Bridging the Nature — Computation dichotomy
Dominic Jones (Netherhall House, London) Sequential Processing in Nature,‘Anything but’ in Scientific ComputationJanuary 2018 14 / 25
Bridging the Nature — Computation dichotomy
Artificial Neural Networks
TensorFlow demo
Dominic Jones (Netherhall House, London) Sequential Processing in Nature,‘Anything but’ in Scientific ComputationJanuary 2018 15 / 25
Bridging the Nature — Computation dichotomy
Properties of ANN
1 Weights are associated with nodes and inter-nodal connections
2 Processes are relatively straight forward - matrix-vector products
and local reductions
3 Its kernel is essentially plastic - number of hidden layers and
number of nodes on each layer govern its form
4 It has somewhat of the ‘it should just work’ essence - a result will
always be produced
Dominic Jones (Netherhall House, London) Sequential Processing in Nature,‘Anything but’ in Scientific ComputationJanuary 2018 16 / 25
Bridging the Nature — Computation dichotomy
Where the analogy breaks down
1 These nodes and these connections are trained for this problem
2 Training is complicated - requires derivative of every output with
respect to every weight
3 Improvement of the weights requires vast resources relative to
evaluating the actual problem
Dominic Jones (Netherhall House, London) Sequential Processing in Nature,‘Anything but’ in Scientific ComputationJanuary 2018 17 / 25
Bridging the Nature — Computation dichotomy
In brief
1 Somehow, the need for feedback (or at least how it is presently
done) appears to lack a natural analogy
2 For protein folding, it is as though all possible solutions (and so
the right one) are ‘known’ at once
3 This is somewhat characteristic of quantum computing, and
moreover, its solution is probably correct
Dominic Jones (Netherhall House, London) Sequential Processing in Nature,‘Anything but’ in Scientific ComputationJanuary 2018 18 / 25
Incidentally...
Incidentally...
Dominic Jones (Netherhall House, London) Sequential Processing in Nature,‘Anything but’ in Scientific ComputationJanuary 2018 19 / 25
Incidentally...
Transpose is like dependency tracing
1 Given a sequence of operations: X(D), Q(X), L(Q),
the Jacobian of Q w.r.t. X can be written as
JQ =


∂Q1
∂X1
· · ·
∂Q1
∂Xn
...
...
...
∂Qm
∂X1
· · ·
∂Qm
∂Xn


=
dQ
dX
2 The derivative of the whole system is the chain of the Jacobians of
each operation, giving
dL
dD
=
dL
dQ
dQ
dX
dX
dD
Dominic Jones (Netherhall House, London) Sequential Processing in Nature,‘Anything but’ in Scientific ComputationJanuary 2018 20 / 25
Incidentally...
Transpose is like dependency tracing
1 The derivative of the system could be built up by inputting
tangents of JX
dL
dDi
=
dL
dQ
dQ
dX
dX
dDi
2 But by turning the tangents approach “inside-out”, taking the
transpose of the derivative of the system for a particular gradient
of JL
dLj
dD
T
=
dX
dD
T
dQ
dX
T dLj
dQ
T
offers an efficient way of relating an output to all of its inputs.
Dominic Jones (Netherhall House, London) Sequential Processing in Nature,‘Anything but’ in Scientific ComputationJanuary 2018 21 / 25
Incidentally...
Give the output a handle on its history
auto eval(A const &a,
B const &b)
{
auto c0 = 3;
auto c1 = 4;
auto t0 = a * b;
auto t1 = c0 + t0;
auto t2 = c1 + t0;
auto r = t1 / t2;
return r;
}
÷
+ +
c0 c1∗ ∗
a b a b
Left branch Right branch
Root
Instead of computing values, construct expression trees (graphs!)
Dominic Jones (Netherhall House, London) Sequential Processing in Nature,‘Anything but’ in Scientific ComputationJanuary 2018 22 / 25
Incidentally...
‘Adjoint’ differentiation
Apply the chain rule and transpose...
t0 = a * b // 1
t1 = c0 + t0 // 2
t2 = c1 + t0 // 3
r = t1 / t2 // 4 // 4
t1_d += (1 / t2) * r_d
t2_d -= (t1 / t2ˆ2) * r_d
// 3
t0_d += t2_d
// 2
t0_d += t1_d
// 1
a_d += b * t0_d
b_d += a * t0_d
Dominic Jones (Netherhall House, London) Sequential Processing in Nature,‘Anything but’ in Scientific ComputationJanuary 2018 23 / 25
Incidentally...
Directed design
Move surface in or out to reduce drag
Dominic Jones (Netherhall House, London) Sequential Processing in Nature,‘Anything but’ in Scientific ComputationJanuary 2018 24 / 25
Incidentally...
In brief
1 There is some analogy with entanglement and wave function
collapsing in classical computation
2 It gives all solutions in one evaluation
3 But, everything really is back-to-front! (and implementing it is a
nightmare!)
Dominic Jones (Netherhall House, London) Sequential Processing in Nature,‘Anything but’ in Scientific ComputationJanuary 2018 25 / 25

Contenu connexe

Similaire à Sequential processing in nature

My data are incomplete and noisy: Information-reduction statistical methods f...
My data are incomplete and noisy: Information-reduction statistical methods f...My data are incomplete and noisy: Information-reduction statistical methods f...
My data are incomplete and noisy: Information-reduction statistical methods f...Umberto Picchini
 
SIAM - Minisymposium on Guaranteed numerical algorithms
SIAM - Minisymposium on Guaranteed numerical algorithmsSIAM - Minisymposium on Guaranteed numerical algorithms
SIAM - Minisymposium on Guaranteed numerical algorithmsJagadeeswaran Rathinavel
 
Slides econometrics-2018-graduate-2
Slides econometrics-2018-graduate-2Slides econometrics-2018-graduate-2
Slides econometrics-2018-graduate-2Arthur Charpentier
 
What is Algorithm - An Overview
What is Algorithm - An OverviewWhat is Algorithm - An Overview
What is Algorithm - An OverviewRamakant Soni
 
Data Structure: Algorithm and analysis
Data Structure: Algorithm and analysisData Structure: Algorithm and analysis
Data Structure: Algorithm and analysisDr. Rajdeep Chatterjee
 
Application of parallel hierarchical matrices and low-rank tensors in spatial...
Application of parallel hierarchical matrices and low-rank tensors in spatial...Application of parallel hierarchical matrices and low-rank tensors in spatial...
Application of parallel hierarchical matrices and low-rank tensors in spatial...Alexander Litvinenko
 
Computer notes - Sorting
Computer notes  - SortingComputer notes  - Sorting
Computer notes - Sortingecomputernotes
 
Physically Unclonable Random Permutations
Physically Unclonable Random PermutationsPhysically Unclonable Random Permutations
Physically Unclonable Random PermutationsRiccardo Bernardini
 
Mit15 082 jf10_lec01
Mit15 082 jf10_lec01Mit15 082 jf10_lec01
Mit15 082 jf10_lec01Saad Liaqat
 
Mm chap08 -_lossy_compression_algorithms
Mm chap08 -_lossy_compression_algorithmsMm chap08 -_lossy_compression_algorithms
Mm chap08 -_lossy_compression_algorithmsEellekwameowusu
 
20130928 automated theorem_proving_harrison
20130928 automated theorem_proving_harrison20130928 automated theorem_proving_harrison
20130928 automated theorem_proving_harrisonComputer Science Club
 
Lesson15 -exponential_growth_and_decay_021_slides
Lesson15  -exponential_growth_and_decay_021_slidesLesson15  -exponential_growth_and_decay_021_slides
Lesson15 -exponential_growth_and_decay_021_slidesMatthew Leingang
 
Lesson 15: Exponential Growth and Decay (Section 021 slides)
Lesson 15: Exponential Growth and Decay (Section 021 slides)Lesson 15: Exponential Growth and Decay (Section 021 slides)
Lesson 15: Exponential Growth and Decay (Section 021 slides)Matthew Leingang
 
Count-Distinct Problem
Count-Distinct ProblemCount-Distinct Problem
Count-Distinct ProblemKai Zhang
 

Similaire à Sequential processing in nature (20)

My data are incomplete and noisy: Information-reduction statistical methods f...
My data are incomplete and noisy: Information-reduction statistical methods f...My data are incomplete and noisy: Information-reduction statistical methods f...
My data are incomplete and noisy: Information-reduction statistical methods f...
 
QMC: Operator Splitting Workshop, Projective Splitting with Forward Steps and...
QMC: Operator Splitting Workshop, Projective Splitting with Forward Steps and...QMC: Operator Splitting Workshop, Projective Splitting with Forward Steps and...
QMC: Operator Splitting Workshop, Projective Splitting with Forward Steps and...
 
SIAM - Minisymposium on Guaranteed numerical algorithms
SIAM - Minisymposium on Guaranteed numerical algorithmsSIAM - Minisymposium on Guaranteed numerical algorithms
SIAM - Minisymposium on Guaranteed numerical algorithms
 
Slides econometrics-2018-graduate-2
Slides econometrics-2018-graduate-2Slides econometrics-2018-graduate-2
Slides econometrics-2018-graduate-2
 
AINL 2016: Strijov
AINL 2016: StrijovAINL 2016: Strijov
AINL 2016: Strijov
 
2.pptx
2.pptx2.pptx
2.pptx
 
What is Algorithm - An Overview
What is Algorithm - An OverviewWhat is Algorithm - An Overview
What is Algorithm - An Overview
 
Data Structure: Algorithm and analysis
Data Structure: Algorithm and analysisData Structure: Algorithm and analysis
Data Structure: Algorithm and analysis
 
Application of parallel hierarchical matrices and low-rank tensors in spatial...
Application of parallel hierarchical matrices and low-rank tensors in spatial...Application of parallel hierarchical matrices and low-rank tensors in spatial...
Application of parallel hierarchical matrices and low-rank tensors in spatial...
 
Computer notes - Sorting
Computer notes  - SortingComputer notes  - Sorting
Computer notes - Sorting
 
Physically Unclonable Random Permutations
Physically Unclonable Random PermutationsPhysically Unclonable Random Permutations
Physically Unclonable Random Permutations
 
Mit15 082 jf10_lec01
Mit15 082 jf10_lec01Mit15 082 jf10_lec01
Mit15 082 jf10_lec01
 
Ch6 slides
Ch6 slidesCh6 slides
Ch6 slides
 
Mm chap08 -_lossy_compression_algorithms
Mm chap08 -_lossy_compression_algorithmsMm chap08 -_lossy_compression_algorithms
Mm chap08 -_lossy_compression_algorithms
 
20130928 automated theorem_proving_harrison
20130928 automated theorem_proving_harrison20130928 automated theorem_proving_harrison
20130928 automated theorem_proving_harrison
 
Logic gates
Logic gatesLogic gates
Logic gates
 
Side 2019 #9
Side 2019 #9Side 2019 #9
Side 2019 #9
 
Lesson15 -exponential_growth_and_decay_021_slides
Lesson15  -exponential_growth_and_decay_021_slidesLesson15  -exponential_growth_and_decay_021_slides
Lesson15 -exponential_growth_and_decay_021_slides
 
Lesson 15: Exponential Growth and Decay (Section 021 slides)
Lesson 15: Exponential Growth and Decay (Section 021 slides)Lesson 15: Exponential Growth and Decay (Section 021 slides)
Lesson 15: Exponential Growth and Decay (Section 021 slides)
 
Count-Distinct Problem
Count-Distinct ProblemCount-Distinct Problem
Count-Distinct Problem
 

Plus de 2idseminar

The importance of being human 3
The importance of being human 3The importance of being human 3
The importance of being human 32idseminar
 
On the axiom of choice
On the axiom of choiceOn the axiom of choice
On the axiom of choice2idseminar
 
Can christian schools continue to teach only about traditional marriage
Can christian schools continue to teach only about traditional marriageCan christian schools continue to teach only about traditional marriage
Can christian schools continue to teach only about traditional marriage2idseminar
 
The unlimited desire IIS2019
The unlimited desire IIS2019The unlimited desire IIS2019
The unlimited desire IIS20192idseminar
 
Causality from outside Time
Causality from outside TimeCausality from outside Time
Causality from outside Time2idseminar
 
Can computers replace teachers?
Can computers replace teachers?Can computers replace teachers?
Can computers replace teachers?2idseminar
 
Future contingents and the Multiverse
Future contingents and the MultiverseFuture contingents and the Multiverse
Future contingents and the Multiverse2idseminar
 
The CRISPR/CAS9 genome editing system and humans
The CRISPR/CAS9 genome editing system and humansThe CRISPR/CAS9 genome editing system and humans
The CRISPR/CAS9 genome editing system and humans2idseminar
 
Achilles, the Tortoise and Quantum Mechanics
Achilles, the Tortoise and Quantum MechanicsAchilles, the Tortoise and Quantum Mechanics
Achilles, the Tortoise and Quantum Mechanics2idseminar
 
Transpostgenderism 5 jean_davidponci
Transpostgenderism 5 jean_davidponciTranspostgenderism 5 jean_davidponci
Transpostgenderism 5 jean_davidponci2idseminar
 
Transhumanism and brain
Transhumanism and brainTranshumanism and brain
Transhumanism and brain2idseminar
 
Transhumanism and brain
Transhumanism and brainTranshumanism and brain
Transhumanism and brain2idseminar
 
Netherhall 2018 m fox
Netherhall 2018 m foxNetherhall 2018 m fox
Netherhall 2018 m fox2idseminar
 
Infinite Chess: winning and draw
Infinite Chess: winning and drawInfinite Chess: winning and draw
Infinite Chess: winning and draw2idseminar
 

Plus de 2idseminar (14)

The importance of being human 3
The importance of being human 3The importance of being human 3
The importance of being human 3
 
On the axiom of choice
On the axiom of choiceOn the axiom of choice
On the axiom of choice
 
Can christian schools continue to teach only about traditional marriage
Can christian schools continue to teach only about traditional marriageCan christian schools continue to teach only about traditional marriage
Can christian schools continue to teach only about traditional marriage
 
The unlimited desire IIS2019
The unlimited desire IIS2019The unlimited desire IIS2019
The unlimited desire IIS2019
 
Causality from outside Time
Causality from outside TimeCausality from outside Time
Causality from outside Time
 
Can computers replace teachers?
Can computers replace teachers?Can computers replace teachers?
Can computers replace teachers?
 
Future contingents and the Multiverse
Future contingents and the MultiverseFuture contingents and the Multiverse
Future contingents and the Multiverse
 
The CRISPR/CAS9 genome editing system and humans
The CRISPR/CAS9 genome editing system and humansThe CRISPR/CAS9 genome editing system and humans
The CRISPR/CAS9 genome editing system and humans
 
Achilles, the Tortoise and Quantum Mechanics
Achilles, the Tortoise and Quantum MechanicsAchilles, the Tortoise and Quantum Mechanics
Achilles, the Tortoise and Quantum Mechanics
 
Transpostgenderism 5 jean_davidponci
Transpostgenderism 5 jean_davidponciTranspostgenderism 5 jean_davidponci
Transpostgenderism 5 jean_davidponci
 
Transhumanism and brain
Transhumanism and brainTranshumanism and brain
Transhumanism and brain
 
Transhumanism and brain
Transhumanism and brainTranshumanism and brain
Transhumanism and brain
 
Netherhall 2018 m fox
Netherhall 2018 m foxNetherhall 2018 m fox
Netherhall 2018 m fox
 
Infinite Chess: winning and draw
Infinite Chess: winning and drawInfinite Chess: winning and draw
Infinite Chess: winning and draw
 

Dernier

GBSN - Microbiology (Unit 2)
GBSN - Microbiology (Unit 2)GBSN - Microbiology (Unit 2)
GBSN - Microbiology (Unit 2)Areesha Ahmad
 
Vip profile Call Girls In Lonavala 9748763073 For Genuine Sex Service At Just...
Vip profile Call Girls In Lonavala 9748763073 For Genuine Sex Service At Just...Vip profile Call Girls In Lonavala 9748763073 For Genuine Sex Service At Just...
Vip profile Call Girls In Lonavala 9748763073 For Genuine Sex Service At Just...Monika Rani
 
PossibleEoarcheanRecordsoftheGeomagneticFieldPreservedintheIsuaSupracrustalBe...
PossibleEoarcheanRecordsoftheGeomagneticFieldPreservedintheIsuaSupracrustalBe...PossibleEoarcheanRecordsoftheGeomagneticFieldPreservedintheIsuaSupracrustalBe...
PossibleEoarcheanRecordsoftheGeomagneticFieldPreservedintheIsuaSupracrustalBe...Sérgio Sacani
 
High Class Escorts in Hyderabad ₹7.5k Pick Up & Drop With Cash Payment 969456...
High Class Escorts in Hyderabad ₹7.5k Pick Up & Drop With Cash Payment 969456...High Class Escorts in Hyderabad ₹7.5k Pick Up & Drop With Cash Payment 969456...
High Class Escorts in Hyderabad ₹7.5k Pick Up & Drop With Cash Payment 969456...chandars293
 
Isotopic evidence of long-lived volcanism on Io
Isotopic evidence of long-lived volcanism on IoIsotopic evidence of long-lived volcanism on Io
Isotopic evidence of long-lived volcanism on IoSérgio Sacani
 
Botany 4th semester series (krishna).pdf
Botany 4th semester series (krishna).pdfBotany 4th semester series (krishna).pdf
Botany 4th semester series (krishna).pdfSumit Kumar yadav
 
Disentangling the origin of chemical differences using GHOST
Disentangling the origin of chemical differences using GHOSTDisentangling the origin of chemical differences using GHOST
Disentangling the origin of chemical differences using GHOSTSérgio Sacani
 
GBSN - Microbiology (Unit 1)
GBSN - Microbiology (Unit 1)GBSN - Microbiology (Unit 1)
GBSN - Microbiology (Unit 1)Areesha Ahmad
 
Nightside clouds and disequilibrium chemistry on the hot Jupiter WASP-43b
Nightside clouds and disequilibrium chemistry on the hot Jupiter WASP-43bNightside clouds and disequilibrium chemistry on the hot Jupiter WASP-43b
Nightside clouds and disequilibrium chemistry on the hot Jupiter WASP-43bSérgio Sacani
 
VIRUSES structure and classification ppt by Dr.Prince C P
VIRUSES structure and classification ppt by Dr.Prince C PVIRUSES structure and classification ppt by Dr.Prince C P
VIRUSES structure and classification ppt by Dr.Prince C PPRINCE C P
 
Botany 4th semester file By Sumit Kumar yadav.pdf
Botany 4th semester file By Sumit Kumar yadav.pdfBotany 4th semester file By Sumit Kumar yadav.pdf
Botany 4th semester file By Sumit Kumar yadav.pdfSumit Kumar yadav
 
Botany krishna series 2nd semester Only Mcq type questions
Botany krishna series 2nd semester Only Mcq type questionsBotany krishna series 2nd semester Only Mcq type questions
Botany krishna series 2nd semester Only Mcq type questionsSumit Kumar yadav
 
Biological Classification BioHack (3).pdf
Biological Classification BioHack (3).pdfBiological Classification BioHack (3).pdf
Biological Classification BioHack (3).pdfmuntazimhurra
 
Kochi ❤CALL GIRL 84099*07087 ❤CALL GIRLS IN Kochi ESCORT SERVICE❤CALL GIRL
Kochi ❤CALL GIRL 84099*07087 ❤CALL GIRLS IN Kochi ESCORT SERVICE❤CALL GIRLKochi ❤CALL GIRL 84099*07087 ❤CALL GIRLS IN Kochi ESCORT SERVICE❤CALL GIRL
Kochi ❤CALL GIRL 84099*07087 ❤CALL GIRLS IN Kochi ESCORT SERVICE❤CALL GIRLkantirani197
 
Labelling Requirements and Label Claims for Dietary Supplements and Recommend...
Labelling Requirements and Label Claims for Dietary Supplements and Recommend...Labelling Requirements and Label Claims for Dietary Supplements and Recommend...
Labelling Requirements and Label Claims for Dietary Supplements and Recommend...Lokesh Kothari
 
Presentation Vikram Lander by Vedansh Gupta.pptx
Presentation Vikram Lander by Vedansh Gupta.pptxPresentation Vikram Lander by Vedansh Gupta.pptx
Presentation Vikram Lander by Vedansh Gupta.pptxgindu3009
 
Formation of low mass protostars and their circumstellar disks
Formation of low mass protostars and their circumstellar disksFormation of low mass protostars and their circumstellar disks
Formation of low mass protostars and their circumstellar disksSérgio Sacani
 
GUIDELINES ON SIMILAR BIOLOGICS Regulatory Requirements for Marketing Authori...
GUIDELINES ON SIMILAR BIOLOGICS Regulatory Requirements for Marketing Authori...GUIDELINES ON SIMILAR BIOLOGICS Regulatory Requirements for Marketing Authori...
GUIDELINES ON SIMILAR BIOLOGICS Regulatory Requirements for Marketing Authori...Lokesh Kothari
 
Pulmonary drug delivery system M.pharm -2nd sem P'ceutics
Pulmonary drug delivery system M.pharm -2nd sem P'ceuticsPulmonary drug delivery system M.pharm -2nd sem P'ceutics
Pulmonary drug delivery system M.pharm -2nd sem P'ceuticssakshisoni2385
 

Dernier (20)

GBSN - Microbiology (Unit 2)
GBSN - Microbiology (Unit 2)GBSN - Microbiology (Unit 2)
GBSN - Microbiology (Unit 2)
 
Vip profile Call Girls In Lonavala 9748763073 For Genuine Sex Service At Just...
Vip profile Call Girls In Lonavala 9748763073 For Genuine Sex Service At Just...Vip profile Call Girls In Lonavala 9748763073 For Genuine Sex Service At Just...
Vip profile Call Girls In Lonavala 9748763073 For Genuine Sex Service At Just...
 
PossibleEoarcheanRecordsoftheGeomagneticFieldPreservedintheIsuaSupracrustalBe...
PossibleEoarcheanRecordsoftheGeomagneticFieldPreservedintheIsuaSupracrustalBe...PossibleEoarcheanRecordsoftheGeomagneticFieldPreservedintheIsuaSupracrustalBe...
PossibleEoarcheanRecordsoftheGeomagneticFieldPreservedintheIsuaSupracrustalBe...
 
High Class Escorts in Hyderabad ₹7.5k Pick Up & Drop With Cash Payment 969456...
High Class Escorts in Hyderabad ₹7.5k Pick Up & Drop With Cash Payment 969456...High Class Escorts in Hyderabad ₹7.5k Pick Up & Drop With Cash Payment 969456...
High Class Escorts in Hyderabad ₹7.5k Pick Up & Drop With Cash Payment 969456...
 
Isotopic evidence of long-lived volcanism on Io
Isotopic evidence of long-lived volcanism on IoIsotopic evidence of long-lived volcanism on Io
Isotopic evidence of long-lived volcanism on Io
 
Botany 4th semester series (krishna).pdf
Botany 4th semester series (krishna).pdfBotany 4th semester series (krishna).pdf
Botany 4th semester series (krishna).pdf
 
Disentangling the origin of chemical differences using GHOST
Disentangling the origin of chemical differences using GHOSTDisentangling the origin of chemical differences using GHOST
Disentangling the origin of chemical differences using GHOST
 
CELL -Structural and Functional unit of life.pdf
CELL -Structural and Functional unit of life.pdfCELL -Structural and Functional unit of life.pdf
CELL -Structural and Functional unit of life.pdf
 
GBSN - Microbiology (Unit 1)
GBSN - Microbiology (Unit 1)GBSN - Microbiology (Unit 1)
GBSN - Microbiology (Unit 1)
 
Nightside clouds and disequilibrium chemistry on the hot Jupiter WASP-43b
Nightside clouds and disequilibrium chemistry on the hot Jupiter WASP-43bNightside clouds and disequilibrium chemistry on the hot Jupiter WASP-43b
Nightside clouds and disequilibrium chemistry on the hot Jupiter WASP-43b
 
VIRUSES structure and classification ppt by Dr.Prince C P
VIRUSES structure and classification ppt by Dr.Prince C PVIRUSES structure and classification ppt by Dr.Prince C P
VIRUSES structure and classification ppt by Dr.Prince C P
 
Botany 4th semester file By Sumit Kumar yadav.pdf
Botany 4th semester file By Sumit Kumar yadav.pdfBotany 4th semester file By Sumit Kumar yadav.pdf
Botany 4th semester file By Sumit Kumar yadav.pdf
 
Botany krishna series 2nd semester Only Mcq type questions
Botany krishna series 2nd semester Only Mcq type questionsBotany krishna series 2nd semester Only Mcq type questions
Botany krishna series 2nd semester Only Mcq type questions
 
Biological Classification BioHack (3).pdf
Biological Classification BioHack (3).pdfBiological Classification BioHack (3).pdf
Biological Classification BioHack (3).pdf
 
Kochi ❤CALL GIRL 84099*07087 ❤CALL GIRLS IN Kochi ESCORT SERVICE❤CALL GIRL
Kochi ❤CALL GIRL 84099*07087 ❤CALL GIRLS IN Kochi ESCORT SERVICE❤CALL GIRLKochi ❤CALL GIRL 84099*07087 ❤CALL GIRLS IN Kochi ESCORT SERVICE❤CALL GIRL
Kochi ❤CALL GIRL 84099*07087 ❤CALL GIRLS IN Kochi ESCORT SERVICE❤CALL GIRL
 
Labelling Requirements and Label Claims for Dietary Supplements and Recommend...
Labelling Requirements and Label Claims for Dietary Supplements and Recommend...Labelling Requirements and Label Claims for Dietary Supplements and Recommend...
Labelling Requirements and Label Claims for Dietary Supplements and Recommend...
 
Presentation Vikram Lander by Vedansh Gupta.pptx
Presentation Vikram Lander by Vedansh Gupta.pptxPresentation Vikram Lander by Vedansh Gupta.pptx
Presentation Vikram Lander by Vedansh Gupta.pptx
 
Formation of low mass protostars and their circumstellar disks
Formation of low mass protostars and their circumstellar disksFormation of low mass protostars and their circumstellar disks
Formation of low mass protostars and their circumstellar disks
 
GUIDELINES ON SIMILAR BIOLOGICS Regulatory Requirements for Marketing Authori...
GUIDELINES ON SIMILAR BIOLOGICS Regulatory Requirements for Marketing Authori...GUIDELINES ON SIMILAR BIOLOGICS Regulatory Requirements for Marketing Authori...
GUIDELINES ON SIMILAR BIOLOGICS Regulatory Requirements for Marketing Authori...
 
Pulmonary drug delivery system M.pharm -2nd sem P'ceutics
Pulmonary drug delivery system M.pharm -2nd sem P'ceuticsPulmonary drug delivery system M.pharm -2nd sem P'ceutics
Pulmonary drug delivery system M.pharm -2nd sem P'ceutics
 

Sequential processing in nature

  • 1. Sequential Processing in Nature, ‘Anything but’ in Scientific Computation an observation Dominic Jones Netherhall House, London January 2018
  • 2. Nature - ‘It should just work’ Nature - ‘It should just work’ Dominic Jones (Netherhall House, London) Sequential Processing in Nature,‘Anything but’ in Scientific ComputationJanuary 2018 2 / 25
  • 3. Nature - ‘It should just work’ Protein folding Synthesising the chain Finding the global minimum Folding occurs in microseconds, but cannot be predicted Dominic Jones (Netherhall House, London) Sequential Processing in Nature,‘Anything but’ in Scientific ComputationJanuary 2018 3 / 25
  • 4. Nature - ‘It should just work’ Abnormal folding Deformed proteins cannot be mended Dominic Jones (Netherhall House, London) Sequential Processing in Nature,‘Anything but’ in Scientific ComputationJanuary 2018 4 / 25
  • 5. Nature - ‘It should just work’ In brief 1 From a chain of amino acids a specific highly complex shape is required 2 There is no apparent error correction in the process (or very little) 3 It is usually successful 4 When it isn’t, Creutzfeldt-Jakob disease, Parkinson’s, Alzheimer’s, etc Dominic Jones (Netherhall House, London) Sequential Processing in Nature,‘Anything but’ in Scientific ComputationJanuary 2018 5 / 25
  • 6. Scientific computation - ‘T = abs(T)’ Scientific computation - ‘T = abs(T)’ Dominic Jones (Netherhall House, London) Sequential Processing in Nature,‘Anything but’ in Scientific ComputationJanuary 2018 6 / 25
  • 7. Scientific computation - ‘T = abs(T)’ Attempts at prediction and projection Numerical approximation Local sensitivities Iterative, discretised and linearised algorithms Dominic Jones (Netherhall House, London) Sequential Processing in Nature,‘Anything but’ in Scientific ComputationJanuary 2018 7 / 25
  • 8. Scientific computation - ‘T = abs(T)’ All starts with a graph (mesh) Resolve change at the small scales Mesh implies graph, implies matrix, implies matrix inversion Dominic Jones (Netherhall House, London) Sequential Processing in Nature,‘Anything but’ in Scientific ComputationJanuary 2018 8 / 25
  • 9. Scientific computation - ‘T = abs(T)’ Matrix inversion: scaling up void inverse(int n, float a[][2*n_max]) { for (int i = 0; i < n; i++) for (int j = n; j < 2*n; j++) a[i][j] = (i == j-n? 1: 0); for (int i = 0; i < n; i++) { // linear float aii = a[i][i]; for (int j = i; j < 2*n; j++) a[i][j] = a[i][j] / aii; for (int j = 0; j < n; j++) { // quadratic! if (i != j) { float aji = a[j][i]; for (int k = 0; k < 2*n; k++) // cubic!! a[j][k] = a[j][k] - aji * a[i][k]; }}}} Dominic Jones (Netherhall House, London) Sequential Processing in Nature,‘Anything but’ in Scientific ComputationJanuary 2018 9 / 25
  • 10. Scientific computation - ‘T = abs(T)’ Computation: slow but compact Geometry and mesh Connectivity graph Dominic Jones (Netherhall House, London) Sequential Processing in Nature,‘Anything but’ in Scientific ComputationJanuary 2018 10 / 25
  • 11. Scientific computation - ‘T = abs(T)’ Topology Representation Symmetric, sparse (48 non-zeros), irregular   a0,0 a0,1 a0,8 a1,0 a1,1 a1,2 a1,6 a2,1 a2,2 a2,3 a3,2 a3,3 a3,4 a4,3 a4,4 a4,5 a4,13 a5,4 a5,5 a5,6 a5,11 a6,1 a6,5 a6,6 a6,7 a7,6 a7,7 a7,8 a7,9 a8,0 a8,7 a8,8 a9,7 a9,9 a9,10 a10,9 a10,10 a10,11 a11,5 a11,10 a11,11 a11,12 a12,11 a12,12 a12,13 a13,4 a13,12 a13,13   Dominic Jones (Netherhall House, London) Sequential Processing in Nature,‘Anything but’ in Scientific ComputationJanuary 2018 11 / 25
  • 12. Scientific computation - ‘T = abs(T)’ Sparsity and Indirection Dense storage: 196 values, 24% efficient Direct access to values float[14][14] A; A[2][3] = 3.142; Compressed row storage: 111 values, 56% efficient Requires indirection to access values (10x slower) int[15] IA = [0, 3, 7, 10, ...]; int[48] JA = [0, 1, 8, 0, 1, 2, 6, 1, 2, 3, ...]; float[48] A; A[JA[IA[2]+2]] = 3.142; // i.e. A[2][3] = 3.142; Dominic Jones (Netherhall House, London) Sequential Processing in Nature,‘Anything but’ in Scientific ComputationJanuary 2018 12 / 25
  • 13. Scientific computation - ‘T = abs(T)’ In brief 1 Many scientific algorithms have graphs and matrices, and perform matrix inversion at their core 2 They are often woefully inefficient due to the requirement to use compressed storage of irregular data 3 The compromise on efficiency is to leverage compact memory footprint 4 In the process of linearising, much of the ‘beauty’ of underlying the mathematics gives way to crude approximation Dominic Jones (Netherhall House, London) Sequential Processing in Nature,‘Anything but’ in Scientific ComputationJanuary 2018 13 / 25
  • 14. Bridging the Nature — Computation dichotomy Bridging the Nature — Computation dichotomy Dominic Jones (Netherhall House, London) Sequential Processing in Nature,‘Anything but’ in Scientific ComputationJanuary 2018 14 / 25
  • 15. Bridging the Nature — Computation dichotomy Artificial Neural Networks TensorFlow demo Dominic Jones (Netherhall House, London) Sequential Processing in Nature,‘Anything but’ in Scientific ComputationJanuary 2018 15 / 25
  • 16. Bridging the Nature — Computation dichotomy Properties of ANN 1 Weights are associated with nodes and inter-nodal connections 2 Processes are relatively straight forward - matrix-vector products and local reductions 3 Its kernel is essentially plastic - number of hidden layers and number of nodes on each layer govern its form 4 It has somewhat of the ‘it should just work’ essence - a result will always be produced Dominic Jones (Netherhall House, London) Sequential Processing in Nature,‘Anything but’ in Scientific ComputationJanuary 2018 16 / 25
  • 17. Bridging the Nature — Computation dichotomy Where the analogy breaks down 1 These nodes and these connections are trained for this problem 2 Training is complicated - requires derivative of every output with respect to every weight 3 Improvement of the weights requires vast resources relative to evaluating the actual problem Dominic Jones (Netherhall House, London) Sequential Processing in Nature,‘Anything but’ in Scientific ComputationJanuary 2018 17 / 25
  • 18. Bridging the Nature — Computation dichotomy In brief 1 Somehow, the need for feedback (or at least how it is presently done) appears to lack a natural analogy 2 For protein folding, it is as though all possible solutions (and so the right one) are ‘known’ at once 3 This is somewhat characteristic of quantum computing, and moreover, its solution is probably correct Dominic Jones (Netherhall House, London) Sequential Processing in Nature,‘Anything but’ in Scientific ComputationJanuary 2018 18 / 25
  • 19. Incidentally... Incidentally... Dominic Jones (Netherhall House, London) Sequential Processing in Nature,‘Anything but’ in Scientific ComputationJanuary 2018 19 / 25
  • 20. Incidentally... Transpose is like dependency tracing 1 Given a sequence of operations: X(D), Q(X), L(Q), the Jacobian of Q w.r.t. X can be written as JQ =   ∂Q1 ∂X1 · · · ∂Q1 ∂Xn ... ... ... ∂Qm ∂X1 · · · ∂Qm ∂Xn   = dQ dX 2 The derivative of the whole system is the chain of the Jacobians of each operation, giving dL dD = dL dQ dQ dX dX dD Dominic Jones (Netherhall House, London) Sequential Processing in Nature,‘Anything but’ in Scientific ComputationJanuary 2018 20 / 25
  • 21. Incidentally... Transpose is like dependency tracing 1 The derivative of the system could be built up by inputting tangents of JX dL dDi = dL dQ dQ dX dX dDi 2 But by turning the tangents approach “inside-out”, taking the transpose of the derivative of the system for a particular gradient of JL dLj dD T = dX dD T dQ dX T dLj dQ T offers an efficient way of relating an output to all of its inputs. Dominic Jones (Netherhall House, London) Sequential Processing in Nature,‘Anything but’ in Scientific ComputationJanuary 2018 21 / 25
  • 22. Incidentally... Give the output a handle on its history auto eval(A const &a, B const &b) { auto c0 = 3; auto c1 = 4; auto t0 = a * b; auto t1 = c0 + t0; auto t2 = c1 + t0; auto r = t1 / t2; return r; } ÷ + + c0 c1∗ ∗ a b a b Left branch Right branch Root Instead of computing values, construct expression trees (graphs!) Dominic Jones (Netherhall House, London) Sequential Processing in Nature,‘Anything but’ in Scientific ComputationJanuary 2018 22 / 25
  • 23. Incidentally... ‘Adjoint’ differentiation Apply the chain rule and transpose... t0 = a * b // 1 t1 = c0 + t0 // 2 t2 = c1 + t0 // 3 r = t1 / t2 // 4 // 4 t1_d += (1 / t2) * r_d t2_d -= (t1 / t2ˆ2) * r_d // 3 t0_d += t2_d // 2 t0_d += t1_d // 1 a_d += b * t0_d b_d += a * t0_d Dominic Jones (Netherhall House, London) Sequential Processing in Nature,‘Anything but’ in Scientific ComputationJanuary 2018 23 / 25
  • 24. Incidentally... Directed design Move surface in or out to reduce drag Dominic Jones (Netherhall House, London) Sequential Processing in Nature,‘Anything but’ in Scientific ComputationJanuary 2018 24 / 25
  • 25. Incidentally... In brief 1 There is some analogy with entanglement and wave function collapsing in classical computation 2 It gives all solutions in one evaluation 3 But, everything really is back-to-front! (and implementing it is a nightmare!) Dominic Jones (Netherhall House, London) Sequential Processing in Nature,‘Anything but’ in Scientific ComputationJanuary 2018 25 / 25