Summary of Benchmarks

January 2023

Strictly Private and Confidential

 SUMMARY OF BENCHMARKS

The following slides are a summary of the benchmarks of the solutions from
Multiverse in real-client, real-life scenarios, and in different industries

Additional details available on request with NDA

Strictly Private and Confidential

 Some benchmarks

Asset Management Trading Quantitative Finance Finance Manufacturing

50% profit increase in real 1M euros in extra profits in 3x faster Option Pricing More accurate and 6x faster 24x faster computer vision
real Portfolio Optimization Forex Trading Credit Scoring prediction for Defect Detection
Optimization

Manufacturing /
Index Tracking Renewable Energy Fraud Detection Automotive Manufacturing
Supply Chain
>2x cumulative returns in 6.2x more accurate in €160K savings (only company $80M/yr savings in Credit 1000x faster in car’s clutch
in Index Tracking Defect Detection beating benchmark) Card Fraud Detection optimization for Quality Control

This list of benchmarks is growing as we speak Strictly Private and Confidential

 QUANTUM &
QUANTUM INSPIRED

PORTFOLIO OPTIMIZATION
C1. Multiverse solver always finds the portfolio with the highest return for a given risk
a given risk matching the state-of-the-art industry solvers for small portfolios*
portfolios*

C 2. M ultiverse so lver find s o p tim al so lutio ns fo r larg e p ro b lem s with

tho usand s of assets with m inim al o verhead while the ind ustry so lvers
sho w a d ram atic increase of runtim e*

C3. Multiverse solver finds better solutions in 24% of cases

than state-of-the-art solvers for large problems*

C4. ALL TOGETHER: Multiverse solver in a real case optimizing the portfolio of a
large bank
*these benchmarks are only applied to static portfolio optimization, given that the classical state-of-the-art solutions used in the market cannot find 4
cannot find any solution in more complex dynamic portfolio optimization problems, where Multiverse Hybrid solver does find optimal solutions.
optimal solutions.
Strictly Private and Confidential
 QUANTUM

C1. BENCHMARK: Multiverse finds the best portfolios*

• Multiverse solver always

finds the same best
portfolios for small
problems as industry
standard tools like Gurobi
solver

• Multiverse solver
outperforms the best-in-
class quantum hybrid
solvers like D-Wave’s

*these benchmarks are only applied to static portfolio optimization, given that the classical state-of-the-art solutions used in the market cannot find 5
any solution in more complex dynamic portfolio optimization problems, where Multiverse Hybrid solver does find optimal solutions.

Strictly Private and Confidential

 QUANTUM

C2. BENCHMARK: Multiverse is faster for large portfolios*

• Multiverse solver finds

optimal solutions for large
portfolio problems with
thousands of assets with
little computational
overhead

• Standard tools show a

dramatic increase in
runtime to reach the same
quality solutions

*these benchmarks are only applied to static portfolio optimization, given that the classical state-of-the-art solutions used in the market cannot find 6
any solution in more complex dynamic portfolio optimization problems, where Multiverse Hybrid solver does find optimal solutions.

Strictly Private and Confidential

 QUANTUM

C3. BENCHMARK: Multiverse

The Multiverse is better in 24% cases for large portfolios*
Hybrid solver
outperforms classical solvers in 24% of cases

• Multiverse solver matches

matches the best solutions
In 76% of cases Multiverse Hybrid
achieves the best solutions solutions found by state-
found by other competitive classical
state-of-the-art solvers in
solvers
in 76% of cases for large
BUT: In large problems with 500
24% of cases
500 assets
Multiverse Hybrid
Classical component sufficient
finds better solutions!
76% • Multiverse solver finds
better solutions than
state-of-the-art solvers in
in the remaining 24%
cases!
*Based on static portfolio optimization problem with 500 assets
*these benchmarks are only applied to static portfolio optimization, given that the classical state-of-the-art solutions used in the market cannot find 7
any solution in more complex dynamic portfolio optimization problems, where Multiverse Hybrid solver does find optimal solutions.

Strictly Private and Confidential

 QUANTUM &
QUANTUM INSPIRED

C4. BENCHMARK: Multiverse finds optimal real dynamic portfolios*

• Multiverse Quantum solution
finds better dynamic
portfolios in real cases than
industry conventional
methods with an 50% profit
increase and 33% volatility
reduction

• These real portfolio problems

are often impossible to solve
for standard methods, while
Multiverse finds always an
optimal solution

*The classical state-of-the-art solutions used in the market cannot find any optimal solution in these complex dynamic portfolio optimization problems, 8
where Multiverse Hybrid solver does find optimal solutions.

Strictly Private and Confidential

 QUANTUM

FOREX EXCHANGE TRADING

F1. Multiverse algorithm gives more accurate predictions and generates

more profit compared to state-of-the-art methods

Strictly Private and Confidential

 QUANTUM

F1. BENCHMARK: Multiverse finds more profitable trading strategy

• Multiverse quantum
Multiverse algorithm outperforms
Quantum
Multiverse strategy
standard trading solutions
Quantum
strategy Multiverse and state-of-the-art
Classical
strategy
€1M
profit classical machine learning*
Client strategy
producing more accurate
Client
strategy predictions for FX rate
movements (EUR/USD)
• Starting from 5M euros
Multiverse trading strategy
generates 1M euros in
extra profit over 5 years!

10
*Multiverse enhanced classical machine learning toolkit.

Strictly Private and Confidential

 QUANTUM
INSPIRED

QUALITY CONTROL IN MANUFACTURING:

ENHANCED QUANTUM INSPIRED
COMPUTER VISION
R1. Multiverse solutions are 3.7x more accurate
than state-of-the-art methods

R2. Multiverse solutions are 9x more efficient

than state-of-the-art methods

11

Strictly Private and Confidential

 QUANTUM
INSPIRED

R1. BENCHMARK: Multiverse catches more production defects

• The detection of defective

pieces in production lines has
a great impact on decreasing
production costs in the
manufacturing industry

• Multiverse Enhanced Deep

Learning is 5.6x better than
industry methods

• Multiverse Tensor Deep

Learning yields a remarkable
6.2x improvement with
higher efficiency

12

Strictly Private and Confidential

 QUANTUM
INSPIRED

R2. BENCHMARK: Multiverse Tensor Deep Learning is highly efficient

• Multiverse Tensor Deep

Learning is 4.5x more
efficient than state-of-the-art
classical Deep Learning
methods

• This increased efficiency

results in optimal memory
usage and faster convergence
to solutions, which is central
in large manufacturing
production lines

13

Strictly Private and Confidential

 QUANTUM

DETECTION OF DEFECTIVE AUTOMOTIVE PARTS:

QUANTUM COMPUTER VISION

N1. Multiverse Quantum Machine Learning solution is 24x faster

than state-of-the-art methods

14

Strictly Private and Confidential

 QUANTUM

N1. BENCHMARK: Multiverse is 24x faster

Nearly the same performance… …but 24x faster training

24
xf
as
ter
• The detection of defected automotive parts is of great impact to improve production lines efficiency and decrease
quality control costs

• Multiverse Quantum Machine Learning solution (preliminary; now being improved) beats classical Machine Learning
methods and matches the level of performances of the best-in-class Deep Learning model by Amazon on the same
dataset. Our solution is 24x faster than Amazon’s advanced solution showing a tremendous quantum potential for
computer vision applications

15

Strictly Private and Confidential

 QUANTUM

OPTIMIZATION OF VEHICLES CLUTCH

CONFIGURATION FOR QUALITY CONTROL

P1. Multiverse solution is 1000x faster and solves >3x larger problems
than the industry benchmark

16

Strictly Private and Confidential

 QUANTUM

P1. BENCHMARK: Multiverse is 1000x faster

• The optimization of clutch configurations during manufacturing is critical for the automotive industry to ensure and
improve the functioning and durability of vehicles’ engines. The difficulty of the optimization due to the high number
of clutch configurations greatly limits the problem size that classical methods can solve

• Multiverse Quantum solution handles 1000x faster the largest problem that can be solved by the client classical solver
with the same quality solution and can solve >3x larger problems, impossible for the industry classical methods

17

Strictly Private and Confidential

 QUANTUM &
QUANTUM INSPIRED

RENEWABLE ENERGY PRODUCTION FORECAST

M1. Multiverse preliminary solution yields lower forecast error

compared to standard industry solutions resulting in more accurate and
more profitable energy forecast predictions

18

Strictly Private and Confidential

 QUANTUM &
QUANTUM INSPIRED

M1. BENCHMARK: Multiverse predicts renewable energy production better

• The forecast of renewable energy production is as hard due to its variability as crucial for the cost-efficient
integration of renewable resources in energy networks such that a 1% yearly forecast error improvement
corresponds to 1M euros savings for the client

• Multiverse Quantum Machine Learning solution provides as preliminary results predictions that are more accurate
than state-of-the-art solutions resulting already in savings for 160K euros (better solutions in progress)
19

Strictly Private and Confidential

 QUANTUM
INSPIRED

OPTION PRICING

A1. Multiverse solutions are 3x faster

than state-of-the-art methods

A2. Multiverse solutions are 3x more efficient

than state-of-the-art methods

20

Strictly Private and Confidential

 QUANTUM
INSPIRED

A1. BENCHMARK: Multiverse is 3x faster

• Bermudan option pricing with
hedging in incomplete real
markets
Time to solution (in steps)

1950
3x • Multiverse algorithms
fa outperform state-of-the-art
st
er methods, like Deep Learning
algorithms

• Multiverse algorithms work

660 even in regimes* where
standard Monte Carlo
methods fail to produce
accurate results
*Solving high dimensional Partial Differential Equations

21
*Solving high dimensional Partial Differential Equations

Strictly Private and Confidential

 QUANTUM
INSPIRED

A2. BENCHMARK: Multiverse is 3x more efficient

• Bermudan option pricing with

hedging in incomplete real
markets
Number of parameters

1057 3x
m • Multiverse algorithms use
or
e much less parameters and are
ef
fic more efficient than state-of-
ie
nt the-art methods

• The use of less parameters is

353
important to provide memory-
efficient solutions for large
problems

*Solving high dimensional Partial Differential Equations

22

Strictly Private and Confidential

 QUANTUM

ENHANCED INDEX TRACKING

G1. Multiverse solutions are able to replicate complex index return behaviours
and improve cumulative returns compared to standard tracking strategies

G2. Multiverse solutions provide reliable and enhanced index tracking with
lower risk compared to standard tracking strategies

23

Strictly Private and Confidential

 QUANTUM

G1. BENCHMARK: Multiverse replicates and improves complex index tracking

• Multiverse Quantum solutions can strictly replicate with 25 assets the index return behaviour of the
Nasdaq Index with 100 assets, at lower risk (blue line)

• Multiverse Quantum solutions can also provide enhanced tracking with a more flexible strategy that allows
to increase the total cumulative return whilst closely replicating daily returns (red line)

24

Strictly Private and Confidential

 QUANTUM

G2. BENCHMARK: Multiverse yields index tracking with less risk

• Multiverse strict tracking solution has nearly exact return tracking behaviour but with lower
risk improving the portfolio return/risk profile (blue line)

• Multiverse enhanced strategy provides a 2x improvement in the return/risk profile* (red line)

25
*return/risk profile measured by the Sharpe ratio: the industry standard way to compare returns of assets with different risks, the higher the better.

Strictly Private and Confidential

 QUANTUM &
QUANTUM INSPIRED

CREDIT SCORING PREDICTION:

FALLEN ANGELS

B1. Multiverse solution catches the same number of credit downgrades

with less false positives and more interpretability
than the industry benchmark

26

Strictly Private and Confidential

 QUANTUM &
QUANTUM INSPIRED

B1. BENCHMARK: Multiverse predicts credit scoring better

• Multiverse Quantum
Machine Learning
catches the same
number of credit
downgrades* with
more precision and less
false positives

• Multiverse solution is
way more interpretable
with much less
classifiers used to
predict the credit
downgrade

27
*these are called Fallen Angels, for example clients/bonds/loans whose solvency is at risk.

Strictly Private and Confidential

Data from Citi QUANTUM

CREDIT CARD FRAUD DETECTION

E1. Multiverse Quantum machine learning catches more credit card

frauds and much faster than state-of-the-art classical machine learning
methods

28

Strictly Private and Confidential

Data from Citi QUANTUM

E1. BENCHMARK: Multiverse catches more credit card frauds

• Multiverse Quantum
machine learning catches
Multiverse
Quantum
more credit card frauds
strategy Multiverse
Classical outperforming state-of-the-
strategy
art classical machine
learning, while also working
up to 100x faster

• This 2% accuracy
improvement can save a
bank up to $80M/year!

29

Strictly Private and Confidential

JOIN US FOR THE QUANTUM LEAP!

Strictly Private and Confidential