analogy from transportation, biology, machine learning

[hyunjimoon]

atoms inspires bits. i'll use this thread to document analogies between #159 and transportation modeling (deterministic and stochastic) and operation control (multi and single stage).

moving observers is current case study as human perspective of case study writer is moving. typology that can collect c and chart it on b would be key for #184

[hyunjimoon]

some input I requested to Cathy for continuous improvement with the goal of creating website like urban or which she agreed to support!

nonresolved q

could we add meaning to intersections?

• mdp (multi,single ctrl); closed sys, goal oriented planning, self-learning
• cumulative diagram (det, stoch mod)
• facility dynamics (stoch mod + multi ctrl)

resolved q

q: multi vs single stage? single stage mostly has linear objective, constraint with optionally integer constraint

[hyunjimoon]

opportunity matrix, replay buffer, expative innovation

Medical - Oncology, Industrial measurement, Video games, Facial and facial expression recognition, Cosmetic surgery, Quality assurance, Inventory management, Dental imaging, Rapid prototyping, Biology - microscopy, Film, i.e., OmniMax, Endoscopy, Quality assurance in packaging, Forensic Computer human interface, Archiving - museums, animals, collectables, Autonomous navigation, Topographical mapping

Robotic feedback, Custom apparel, hairstyles, clothes, Sports, Architectural planning, Virtual meetings, Cloning, Consumer still imaging, Consumer video

feasibility and optimality cuts

burst and leak detection was their goal for commercialization, they did list "all the other possible product/market fits" (creating an Opportunity Matrix):

1. Pipe burst and leak detection, 2. Water demand prediction, 3. Pump optimization, 4. Water utility operations simulations, 5. Water pipe pressure management, 6. Water quality management

"replay buffer" in sequential decision making (de-correlates samples § Increased diversity in data; less likely that the data overall is bad for learning; Approximate QL as "Pseudo"-Gradient Descent; Intuition: Unlike in supervised learning, in RL, "agent collects its own data". If that data is bad, then the result is bad too (and may make future collection of data even worse). Also: Correlated samples. Violates i.i.d. assumption in supervised learning.)

[hyunjimoon]

lane changing behavior ~ pivot (asymmetric makes model interesting) regulation: overtaking on the right is illegal in europe

sum of the utility increases as a whole (with my lane change)
IDM is longitudinal (a, b accel and deccel) and lane changing is latitudinal so mixing them becomes more complex, although both are microscopic

[hyunjimoon]

	car_traffic_mdp.pdf	startup traffic
objective	Minimize N cars' waiting time	Minimize N technologies' waiting time
behavior	Lane changing	Pivot
states	Discretize incoming lanes into cells to represent vehicle presence	Discretize technology development stages into phases to represent progress
Actions	Four actions corresponding to traffic light phases for different directions	Four actions corresponding to strategic shifts: product development, market expansion, scaling operations, and partnership or investment seeking
State Transitions	Use SUMO to simulate traffic signal environment	Use a business simulation model to represent the market and operational environment
Rewards	Change in cumulative stopped time between actions	Change in cumulative delay in technology deployment or market entry
agent	Car	Startup
environment	Road	Market

This analogy captures the essence of applying MDP principles from traffic control to startup technology development, focusing on strategies that minimize delays and optimize performance in a dynamic market environment.

[hyunjimoon]

Table 4: Goal and Belief-Based Operations of Entrepreneur and Biology

	Entrepreneur Biology	Biology
AGENT in NAIL, SCALE, SAIL phase with belief $b_t$	Described stages: YOUNG_AGE, WORKING_AGE, OLD_AGE	LIFE STAGES: Juvenile, Adult, Senescent. Juveniles are growing and learning, dependent on parental care, exploring their environment with minimal resources. Adults are mature, reproducing, and contributing to their population, often facing time and resource allocation challenges. Senescent individuals are in their declining phase, facing decreased energy and health issues, reflecting on their life's achievements.
VECHILE_AGENT in NAIL, SCALE, SAIL phase	NAIL_STAGE, SCALE_STAGE, SAIL_STAGE definitions	DEVELOPMENT PHASES: Seedling/Germination, Growth/Reproduction, Maturity/Senescence. Seedling phase focuses on initial establishment and survival. Growth phase is about expansion, reproduction, and solidifying ecological presence. Maturity phase is about optimizing survival strategies and preparing for the next generation.
VECHILE_AGENT_GOAL $g$	Create and grow sustainable business, innovate solutions, achieve market success.	BIOLOGICAL GOAL: To survive, reproduce, and ensure the survival of offspring, thus contributing to the genetic continuity and evolution of the species.
innate, how to acquire	nothing is innate, acquire by buy or make	INNATE AND ACQUIRED: Genetic inheritance is innate, providing the initial setup for survival. Acquired traits and adaptations come from interaction with the environment, learning, and physiological changes.
input	people, revenue, investment, solution, candidate of technologies, sampled need	RESOURCES: Sunlight, water, nutrients, genetic material. Inputs vary by species but fundamentally include elements necessary for growth and reproduction.
output	product, value, cost, need solution pair	OUTCOMES: Biomass, offspring, ecological contributions (e.g., oxygen production by plants), and genetic variations.
Relation between MACRO_ACT, MICRO_ACT, NANO_ACT	Use of algorithmic learning from startup cases	BIOLOGICAL PROCESSES: Evolutionary adaptations, individual growth, and cellular functions. Macro acts could be species evolution, micro acts could be individual development, and nano acts could be cellular or genetic processes.
growth structure	startup's growth tree structure	GROWTH STRUCTURE: Phylogenetic tree structure representing evolutionary relationships and developmental stages within an organism's life cycle.
MACRO_ACT $a^a_t$	Strategy and orientation choices in the entrepreneurial ecosystem	ECOLOGICAL STRATEGY: Survival strategies that species use to adapt to their environments, such as niche adaptation, competition, and symbiosis.
MICRO_ACT $a^i_t$	Resource allocation decisions	DEVELOPMENTAL DECISIONS: How an organism allocates its resources to growth, reproduction, and maintenance.
NANO_ACT $a^n_t$	Tools and actions at the granular level	CELLULAR AND MOLECULAR ACTIONS: Biological processes at the cellular or molecular level, such as gene expression, protein synthesis, and metabolic pathways.
ENVIRONMENT	Industry and societal contexts	ECOSYSTEM: The natural environment in which organisms live, including biotic and abiotic factors.
COMPETE and COLLABORATE relation in EVOLVE	Entrepreneurial dynamics of competition and collaboration	ECOLOGICAL INTERACTIONS: Predation, competition, mutualism, and commensalism among species within an ecosystem.
EVOLVE_co_opt_NAIL, EVOLVE_co_adapt_SCALE, EVOLVE_co_opt_SAIL, EVOLVE_co_adapt_SAIL	Stages of startup evolution	EVOLUTIONARY ADAPTATIONS: Stages in an organism's or species' adaptation to its environment, including genetic mutations, natural selection, and co-evolution with other species.
tradeoff objectives	Balancing between solution and need implementation	RESOURCE ALLOCATION TRADEOFFS: Balancing between growth, reproduction, and survival in changing environments.

This systematic approach helps translate the entrepreneurial concepts into biological terms, drawing parallels between the growth and operational strategies of

[hyunjimoon]

1. analogy from what could go wrong for startup gene

DNA is in every cell of our body. It's our hard drive. It's where the entire code that determines how we leave the store that code gets translated. It's transcribed into RNA into pre messenger RNA is that chain or were there that pre mRNA gets spliced into our messenger RNA gets exported out of the nucleus and into the cytoplasm, where ribosomes will read that messenger RNA and make it into a probe protein into an amino acid chain that then folds in a very specific way to give rise to a protein.
Of course proteins are at the business end of life. Proteins are left to function or compressed tissue or compressed cells are communicating between organs and everything else. Disease can occur at any one of this steps. You can have a defect in the DNA. There can be an issue in transcription or splicing or even transport of r&d.
There can be a challenge during the translation of the RNA and there can be this folding of the process. And at every one of those stages if anything goes wrong disease arises.

Category	Car	Pharmaceutical
Features	Defined by car maker	Defined by clinical endpoint in trial
Price	Defined by market	Reimbursement defined by efficacy outcomes, reviewed by CMS
Personal Choice	Defined by consumer	Prescribed by physician based on clinical endpoints

In this table:

• "Features" for a car are determined by the manufacturer who designs and builds the car, while the features of a pharmaceutical product are determined by the results of clinical trials, specifically the clinical endpoints which measure the outcomes of interest in the trial.

• "Price" for a car is set by the market, which could include factors like supply and demand, competition, and production costs. For pharmaceuticals, the price is effectively the reimbursement level, which is decided based on the efficacy outcomes of the drug as assessed by the Centers for Medicare & Medicaid Services (CMS).

• "Personal Choice" with a car refers to the decisions made by the consumer on what car to purchase based on personal preferences, needs, and financial means. In contrast, personal choice with pharmaceuticals is generally in the hands of the prescribing physician, who chooses which drug to prescribe based on the clinical endpoints achieved by the drug in its trials.

3. need to rerun clinical trial due to risk benefit ratio differ (viagra for heart vs 정력 has different target RB profile)

4. hierarchical concept
   symptomic, pathology, cause of disease
   metabolic, protein, gene

5. tumor size is good biomarker for lung and breast cancer but not for

6. two uc: feasibility (can we make product) and desirability (can we sell pd)

7. to get paid (regulatory, investment), we need to show economic benefit in two: approved + paid for. sometimes restricting population (homogeneity) to increase efficacy probability is worthwhile

8. generic need to prove equivalence (chemical composition), new drug need 5-10k and need to prove improved efficacy or safety + superiority + risk/beneift

9. what body does to the drug (pharmacokinetics;distribtuion meabolishm) VS what drugs does to the body (pharmacodynamics;efficay )

[hyunjimoon]

analogy to transfer knowledge from parallel chain of posterior search to ent.search, implicit prior with bayesian simulation to ent.prior formation (strong to weak), technical debt to preventative maintenance of capability

Margossian_Modernizing Markov chains Monte Carlo for Scientific and Bayesian Modeling.pdf
burkner24_genBMImpPrior.pdf
hidden-technical-debt-in-machine-learning-systems-Paper.pdf

Guru & Paper	Machine Learning Concept	Entrepreneurial Application
Charles Margossian - Modernizing Markov chains Monte Carlo for Scientific and Bayesian Modeling	Use short parallel chains for complex posterior geometries	Run multiple small-scale experiments simultaneously to discover diverse potential paths to success. Each experiment has the chance to uncover unique market niches or value propositions.
	High-dimensional parameter spaces benefit from short parallel chains	In complex business environments, conduct several quick, focused tests to reveal insights that a single, prolonged strategy might overlook.
	Utilize parallel computing resources	Leverage a diverse team or network of mentors/advisors to simultaneously test different aspects of a business model.
	Use nested R̂ (n̂R) for convergence diagnostics	Employ nuanced metrics to evaluate the success of multiple small experiments, rather than relying solely on traditional KPIs.
Paul-Christian Bürkner - Generative Bayesian Modeling with Implicit Priors	Use implicit priors	Incorporate small amounts of real or simulated data to create more stable and realistic priors for your business model. This can prevent unreasonable outcomes in business simulations and planning.
	Improve simulation stability	Use implicit priors to stabilize business simulations, especially for complex, multi-faceted business models. This leads to more reliable outcomes in strategic planning.
	Incorporate real-world data	Use small portions of actual market data or initial experiment results to inform your priors. This grounds business assumptions in reality.
	Enhance model generalizability	Consider how your business model might perform across different contexts or industries. Use implicit priors to create more flexible models.
	Simulation-based calibration	Integrate your business model with simulation-based calibration techniques. This helps validate your model across different business scenarios.
D. Sculley - Hidden Technical Debt in Machine Learning Systems	Boundary Erosion	Clearly define and regularly review the core purpose and scope of your startup. Implement a formal process for evaluating new opportunities or product features. Regularly assess if current activities align with the company's mission and strategy.
	Entanglement	Design modular business processes that can be improved independently. Avoid creating unnecessary dependencies between different parts of the business. Consider how new products/services might impact existing offerings.
	Hidden Feedback Loops	Regularly analyze how different parts of your business impact each other. Implement systems to track unexpected consequences of business decisions. Be cautious of making multiple significant changes simultaneously.
	Undeclared Consumers	Maintain a clear inventory of who uses your product/service and how. Implement formal processes for tracking and approving new use cases. Regularly communicate with customers to understand evolving needs.
	Data Dependencies	Carefully manage and document all data sources used in decision-making. Regularly assess the reliability and relevance of your data sources. Implement versioning for critical data to allow for rollbacks if needed.
	Configuration Issues	Document all key business processes and decision-making frameworks. Implement version control for important business documents and strategies. Regularly review and update standard operating procedures.
	Changes in the External World	Implement a formal process for environmental scanning and trend analysis. Regularly reassess market conditions and adjust strategies accordingly. Build flexibility into long-term plans to accommodate external changes.
	System-Level Anti-Patterns	Regularly audit business processes for inefficiencies or redundancies. Avoid quick fixes that might create long-term problems. Foster a culture that values sustainable, scalable solutions over short-term gains.

with claude.Applying Machine Learning to Entrepreneurial Search

[hyunjimoon]

Prof. Prateek Bansal from the National University of Singapore

Harnessing Household Travel Survey with Smart Card Data to Generate Spatiotemporally Heterogeneous Activity Schedules for Transit Users

uses transit data from seoul