As number of asset, constraint, and market parameters increase, classical portfolio optimization models becomes quite complicated to solve. Financial institutions should manage multiple objectives such as expected return, risk and return, liquidity, cost of trading, regulations, and ESG preferences under uncertain markets. Therefore the problem is a multi-objective optimization one.
Quantum-enhanced portfolio optimization is the combination of artificial intelligence (AI), classical high-performance computing (HPC), and quantum optimization algorithms that can help explore larger solution spaces faster for some optimization problem classes.
Quantum is expected to augment the AI tools, rather than replace them as a supplement for the high computationally demanding optimization problems when quantum computers become a mature technology. Core Capabilities: Portfolio Optimization, Risk management, Capital Allocation, Trading Optimization, and derivative portfolio optimization. The Benefits of Quantum Enhanced portfolio Optimization: Better exploration in the complexity space. Managing complex portfolio sizes.
Faster portfolio scenario analysis and estimation.
Improve portfolio rebalancing capabilities and adaptability with AI. Potential speed benefits on certain optimization and simulation tasks (once fault tolerant QCs are ready). Limitations: Noise in the systems, immature quantum hardware.
Application that utilize the current QCs only with limitations for the financial use case, or require robust fault tolerant QCs. Financial specific workflows, governance and regulatory compliance needs integration, explainable AI in order to increase trust in predictions for the use case. Market and asset price predictions are expected to work better from AI not QCs.
Strategic Vision: The future will be around hybrid intelligence.
Standard classical compute will dominate most of the standard analyses and transactions. The new era with the usage of AI will bring deeper insights and the capability to adapt to market volatility, and QCs will help augment computing capability for specific computationally challenging parts of the portfolio optimization and simulation processes when becoming mature. Financial Institutions that invest early with the appropriate tools (QCs aware algorithms, AI Analytics, a better computed infrastructure, governance) will be ready to take benefit from the future quantum enabled world.
Another interesting deep dive! That statement at the end ..."for now, out-of-order execution is still too complex to pull off at sub-1W targets, even with modern process nodes." brought up this question: What is the approximate power envelope of Apple's efficiency cores? Those have been out-of-order designs for several generations.
It technically belongs to the 7-series, since it's meant to serve as the "big" core in a big.little setup. Arm changed the naming scheme with A72, but A72 is the direct successor to A57
"A55 then served as a little core partner to A75, A76, and A78"
A55 served as partner to many other Arm cores. It is still probably the most popular Arm core in the world.
DSU friends list:
* A75
* A76
* A77 (previously felt left out xD)
* A78
* A710
* X1
* X2
Exotic, automotive friends:
* A65 (really weird, supported SMT)
* many Axx (AE variants)
Regarding implemetation with A710 and A55 - this seems ridiculous because there were A510 already available at the time.
However, while A710 supported AArch32 at EL0, problematic A510(v1) supported AArch64 only. Meaning bye bye for 32-bit Android apps on the little cores (still quite popular at that time)
Only after complaints from Qualcomm and Mediatek, Arm released A510(v2) which supported AArch32 at EL0.
So, yeah A55 is probably the most popular "Application" Arm core in the world.
With the fact that A55 mostly remains the same in the backend v. A53, I can see how the A55 is what the A53 could've been if its memory performance (specifically bandwidth) was a lot better. Like nearly 8 GB/s from an A55 isn't what I would expect, but it does exist...
Great closing point on the tradeoffs between in order and out of order for low power at the end. It would have been great to see the power for the A53, A55, A73 geekbench scores at the end, in case you have those measurements from when you tested.
As number of asset, constraint, and market parameters increase, classical portfolio optimization models becomes quite complicated to solve. Financial institutions should manage multiple objectives such as expected return, risk and return, liquidity, cost of trading, regulations, and ESG preferences under uncertain markets. Therefore the problem is a multi-objective optimization one.
Quantum-enhanced portfolio optimization is the combination of artificial intelligence (AI), classical high-performance computing (HPC), and quantum optimization algorithms that can help explore larger solution spaces faster for some optimization problem classes.
Quantum is expected to augment the AI tools, rather than replace them as a supplement for the high computationally demanding optimization problems when quantum computers become a mature technology. Core Capabilities: Portfolio Optimization, Risk management, Capital Allocation, Trading Optimization, and derivative portfolio optimization. The Benefits of Quantum Enhanced portfolio Optimization: Better exploration in the complexity space. Managing complex portfolio sizes.
Faster portfolio scenario analysis and estimation.
Improve portfolio rebalancing capabilities and adaptability with AI. Potential speed benefits on certain optimization and simulation tasks (once fault tolerant QCs are ready). Limitations: Noise in the systems, immature quantum hardware.
Application that utilize the current QCs only with limitations for the financial use case, or require robust fault tolerant QCs. Financial specific workflows, governance and regulatory compliance needs integration, explainable AI in order to increase trust in predictions for the use case. Market and asset price predictions are expected to work better from AI not QCs.
Strategic Vision: The future will be around hybrid intelligence.
Standard classical compute will dominate most of the standard analyses and transactions. The new era with the usage of AI will bring deeper insights and the capability to adapt to market volatility, and QCs will help augment computing capability for specific computationally challenging parts of the portfolio optimization and simulation processes when becoming mature. Financial Institutions that invest early with the appropriate tools (QCs aware algorithms, AI Analytics, a better computed infrastructure, governance) will be ready to take benefit from the future quantum enabled world.
Another interesting deep dive! That statement at the end ..."for now, out-of-order execution is still too complex to pull off at sub-1W targets, even with modern process nodes." brought up this question: What is the approximate power envelope of Apple's efficiency cores? Those have been out-of-order designs for several generations.
nit: a57 is probably not a 7 series core :)
It technically belongs to the 7-series, since it's meant to serve as the "big" core in a big.little setup. Arm changed the naming scheme with A72, but A72 is the direct successor to A57
Great analysis, as always.
"A55 then served as a little core partner to A75, A76, and A78"
A55 served as partner to many other Arm cores. It is still probably the most popular Arm core in the world.
DSU friends list:
* A75
* A76
* A77 (previously felt left out xD)
* A78
* A710
* X1
* X2
Exotic, automotive friends:
* A65 (really weird, supported SMT)
* many Axx (AE variants)
Regarding implemetation with A710 and A55 - this seems ridiculous because there were A510 already available at the time.
However, while A710 supported AArch32 at EL0, problematic A510(v1) supported AArch64 only. Meaning bye bye for 32-bit Android apps on the little cores (still quite popular at that time)
Only after complaints from Qualcomm and Mediatek, Arm released A510(v2) which supported AArch32 at EL0.
So, yeah A55 is probably the most popular "Application" Arm core in the world.
For the Mediatek SBC, which distro/kernel version of linux was used?
Ubuntu?
Ubuntu 22.04.5 LTS, 5.15.0-1041-mtk
With the fact that A55 mostly remains the same in the backend v. A53, I can see how the A55 is what the A53 could've been if its memory performance (specifically bandwidth) was a lot better. Like nearly 8 GB/s from an A55 isn't what I would expect, but it does exist...
Great closing point on the tradeoffs between in order and out of order for low power at the end. It would have been great to see the power for the A53, A55, A73 geekbench scores at the end, in case you have those measurements from when you tested.
Difficult to measure power because I'd have to do that at the outlet for the A53/A73 SBC, and at the USB-C port for the A55 one