arg min

The 'arg min' Substack explores the intricacies of machine learning, statistical methods, and the impact of technology on decision making. It delves into the history, challenges, and philosophical questions surrounding artificial intelligence, emphasizing the importance of optimization, data analysis, and the balance between theory and application in scientific advancements.

Machine Learning Statistical Methods Artificial Intelligence Data Analysis Scientific Communication Optimization Techniques History of Technology Philosophy of Science

The hottest Substack posts of arg min

And their main takeaways

Hidden Foundations

138 implied HN points • 15 Sep 23
  1. Machine learning algorithms rely on representing data as abstract mathematical vectors
  2. Images can be represented as arrays on a computer using spatial filtering techniques
  3. Prior knowledge and understanding of data are crucial in generating meaningful features for machine learning

Are there always trade offs?

158 implied HN points • 07 Aug 23
  1. Optimization algorithms are good at finding minimum values for single objectives under constraints.
  2. In cases of multiple objectives, weighing them can lead to a complex surface of potential solutions called the Pareto Frontier.
  3. Optimization in some fields involves finding a balance between different objectives, like fitting a model to data while keeping it simple.

Evil Andrew Huberman’s Fitness Program Can’t Fail

158 implied HN points • 25 Jul 23
  1. Fitness programs based on progressive overload work effectively for strength training
  2. Patterns and logic guide strength training programs, progressing through repetitive cycles
  3. Beware of false claims of 'science-based' fitness advice by influencers, focus on the core idea of progressively overloading the body

Saturday food blogging

138 implied HN points • 26 Aug 23
  1. Farmers' markets in California offer seasonal and fresh produce
  2. Late summer is a great time to enjoy the best stone fruit and tomatoes
  3. A simple and tasty salad recipe using fresh market ingredients, perfect for summer

If you can’t beat em, join em.

99 implied HN points • 05 Nov 23
  1. Drummers have adapted to sound like drum machines.
  2. Technology has revolutionized the way drumming is approached and performed.
  3. Incorporating drum and bass influences has influenced modern drumming styles.
Get a weekly roundup of the best Substack posts, by hacker news affinity:

We need an instrument to take a measurement

99 implied HN points • 03 Nov 23
  1. LATE is a clever way to estimate treatment effects indirectly
  2. Applying LATE requires valid assumptions about causal graphs
  3. Using LATE as an instrumental variable in observational studies can lead to misleading results

Three Causal Graphs

99 implied HN points • 31 Oct 23
  1. Observational studies have limitations in determining causality due to confounding variables like socioeconomic status.
  2. Collider bias can occur when variables become correlated after selecting samples based on extreme cases.
  3. Mediators play a role in causal inference and can be mistaken for confounders, affecting estimates of effects.

Look before you leap

99 implied HN points • 19 Oct 23
  1. Good decisions are based on good measurements.
  2. Balancing true positive rate and false positive rate is crucial in decision-making.
  3. ROC curves provide insight into the limits of noisy measurements and the quality of decisions.

The fake rigor of the mighty dollar

99 implied HN points • 18 Oct 23
  1. Uncertainty in modeling can impact decision-making
  2. Quantifying costs and benefits is challenging due to uncertainty
  3. Utilitarianism and similar philosophies may not account for uncertainties in decision-making

A month of milliblogging

138 implied HN points • 30 Jul 23
  1. The author found value in daily writing practice.
  2. Writing consistently helped the author develop new writing skills.
  3. Future plans include exploring topics on machine learning and teaching with a critical eye.

From Quantity Comes Quality

138 implied HN points • 16 Jul 23
  1. The idea of 'From Quantity Comes Quality' emphasizes creating a lot of content to find the best parts.
  2. Applying the principle of 'From Quantity Comes Quality' can be beneficial in various creative processes, like making music or writing.
  3. Having a structured approach and sticking to self-imposed rules can help overcome creative blocks and enhance productivity.

Soylent 1945

119 implied HN points • 16 Aug 23
  1. The concept of finding the optimal diet is complex and unrealistic.
  2. The 'optimal diet problem' can be simplified into a linear programming exercise.
  3. Attempting to optimize diets based on linear programs may not consider important factors like food variety, palatability, and social value.

A week of millibloging

138 implied HN points • 08 Jul 23
  1. The author finds milliblogging on Substack more rewarding than tweeting.
  2. Milliblogging allows for more complete sentences, context, and civil discussions compared to Twitter.
  3. The author plans to experiment with daily posts in July to develop half-baked thoughts into a cohesive narrative.

Difficult Statistics

138 implied HN points • 07 Jul 23
  1. There are different viewpoints in statistics: Bayesians and Frequentists.
  2. David Freedman emphasized the importance of good data in statistics.
  3. Critiques of statistical practices can help improve methodology.

The suboptimal life.

119 implied HN points • 09 Aug 23
  1. Optimization is not always necessary, and it's okay if we're not optimizing everything.
  2. Inverse Optimization can be challenging, as not everything in reality optimizes something.
  3. The idea that everything is optimization may not always be accurate, as seen in fields like economics.

Some simple plans

79 implied HN points • 08 Nov 23
  1. Stochastic optimization problems can be solved with stochastic gradient descent.
  2. Policy optimization examples include portfolio allocation, demand forecasting, and location optimization.
  3. These problems involve making a single decision based on stochastic data and a random model of uncertainty.

What's wrong with the case study?

119 implied HN points • 19 Jul 23
  1. The case studies on penicillin's effectiveness provided valuable insights into treatment outcomes and dosing regimens.
  2. Case studies are often dismissed as weak evidence but can offer detailed and informative narratives for medical research.
  3. Observations from case studies can help complement traditional trial data and expand our understanding of treatments.

One hundred years of experimentude

119 implied HN points • 12 Jul 23
  1. 2023 marks the 100th anniversary of the scientific method.
  2. Neyman's paper introduced randomized controlled experiments as a form of measurement.
  3. Neyman's proposed method for clinical trials revolutionized how treatment efficacy is evaluated.

How to win an argument with poetry

119 implied HN points • 09 Jul 23
  1. One can use poetry as a powerful tool in winning arguments.
  2. The concept of beauty in science can be broad and complex.
  3. Even if not religious or a poetry enthusiast, impactful poems can still resonate.

From proportion to causation

119 implied HN points • 06 Jul 23
  1. Regression can be confusing, and that's okay.
  2. Conditional probability is a useful tool to understand why certain things happen.
  3. Logistic regression can help model relationships between attributes and outcomes, but its application may not always reflect real-world decision-making processes.

Layers upon layers of complexity

99 implied HN points • 28 Jul 23
  1. Training for sports involves complex systems and interactions
  2. Fitness-fatigue models may not capture the full complexity of training needs
  3. Periodization is key for planning training programs effectively

The maddening nonlinear multiverse

79 implied HN points • 19 Sep 23
  1. Nonlinear prediction involves creating more features with simple nonlinear rules.
  2. Different methods of building nonlinear predictors essentially do the same thing - mapping features into a higher dimensional space.
  3. Various approaches exist to find efficient and effective basis functions for nonlinear prediction.

Learning from dynamics

99 implied HN points • 21 Jul 23
  1. N-of-1 experiments provide more detailed information about drug effects over time than traditional trials.
  2. Experiment design in engineering aims to model how systems respond to inputs.
  3. Utilizing random inputs in system identification can efficiently estimate parameters for linear and nonlinear systems.

Regulations, Rules, and RCTs.

99 implied HN points • 13 Jul 23
  1. Randomized trials are used to measure differences in outcomes between treatments.
  2. Randomized trials help make decisions by estimating treatment effects and values, informing cost-benefit analysis.
  3. Randomized experiments are more impactful in regulatory mechanisms than for scientific discoveries.

Saying the same thing in different ways

99 implied HN points • 11 Jul 23
  1. P-values and confidence intervals convey similar information but in different ways.
  2. P-values are often misunderstood, but they usually refer to the probability of a specific event given certain conditions.
  3. Confidence intervals and p-values are more reliable when dealing with normally distributed data.

I have no confidence in Confidence Intervals

99 implied HN points • 10 Jul 23
  1. The author questions the use of Confidence Intervals in statistics.
  2. Confidence Intervals are based on the assumption of normal distribution and can be limiting in making certain inferences.
  3. In measurement, random errors serve as a model to gauge precision, helping improve the accuracy of devices.

The inductive leap

59 implied HN points • 16 Oct 23
  1. When making decisions based on data-driven patterns, consider modeling past patterns to predict future outcomes.
  2. Inductive leaps involve using statistics from past events to make educated guesses for decision-making.
  3. It's important to connect actions with potential outcomes under uncertainty through optimization balancing costs and benefits.

There is no optimum

79 implied HN points • 01 Aug 23
  1. People in the fitness industry often sell programs to make a living, leading to a variety of options.
  2. Different optimization methods can work in different contexts, so there's no one 'right' algorithm for maximizing objectives.
  3. The effectiveness of a fitness program depends on the individual's experience level and requires personalization.

Gaming the algorithm

79 implied HN points • 23 Jul 23
  1. Experiment with using Spotify's Daily Mix feature to discover new music in a specific genre
  2. Consider creating a custom playlist by copying and refining Daily Mixes to tailor music recommendations to your preferences
  3. Utilize AI-generated playlists to find new artists and genres you might enjoy

Is anybody alive in here?

79 implied HN points • 15 Jul 23
  1. The author reflects on the 40th anniversary of becoming a music fan with The Police's album 'Synchronicity'.
  2. The A-side of 'Synchronicity' is described as arty prog-rock, while the B-side is seen as Sting's best solo record with pop masterpieces.
  3. The Police's unique sound, dynamic drumming, creative guitar work, and nerdy music approach contributed to their success despite internal conflicts.

A Dappled World of Probability

79 implied HN points • 05 Jul 23
  1. Probability is a property of mathematical models, useful when it corresponds well with the natural system it describes.
  2. Examples where probability can be applied include games of chance, randomized algorithms, and randomized controlled trials.
  3. There are different views of probability, including mathematical axioms, Frequentist and Bayesian perspectives, each useful in specific contexts but switching between them can cause confusion.

Superforecasters are not superheroes

79 implied HN points • 03 Jul 23
  1. Superforecasters aim to predict better than experts, but their methods may be flawed.
  2. Using subjective probabilities to answer yes-no questions may not be reliable for accurate predictions.
  3. Superforecasters may struggle with predicting rare events and their accuracy decreases over time.

Deciding not to decide

59 implied HN points • 18 Aug 23
  1. Naturalistic decision making involves more than just solving complex optimization problems.
  2. Human decision making often relies on pattern recognition, different from machine-based recognition.
  3. Automating human decision-making remains a challenging task due to the unique nature of how humans make decisions.

Bean counting

59 implied HN points • 17 Aug 23
  1. Started a high-protein diet for better recovery after Coach's recommendation
  2. Logging food helped discover hidden calories and create a calorie deficit
  3. Realized US food marketing can confuse healthy eating efforts

Healthcare and the AI dystopia

59 implied HN points • 11 Aug 23
  1. Healthcare decisions are increasingly guided by AI tools and algorithms.
  2. Evaluation of algorithmic decision systems in healthcare is challenging and lacks clear methods.
  3. Medical guidelines and recommendations have a significant impact on patient care outcomes.

Extremely Online

39 implied HN points • 09 Nov 23
  1. Regret measures how well we use observations to make decisions.
  2. In the multiarmed bandit problem, the optimal strategy involves a balance of exploration and exploitation.
  3. Considering the impact of actions on future outcomes is crucial in dynamic policy optimization.

And now we come to the feats of strength

59 implied HN points • 06 Aug 23
  1. Strongman is a sport where athletes move odd and heavy objects in a short time.
  2. The sport of Strongman has roots in the circus and television, aiming to showcase feats of strength.
  3. Strongman competitions feature events like lifting giant stones, deadlifting unusual objects, and carrying heavy loads in creative ways.

It's All in My Mind

59 implied HN points • 03 Aug 23
  1. Physical therapy uses progressive overload to heal injuries
  2. Belief in physical therapy effectiveness can be influenced by misconceptions about healing processes
  3. Physical therapy can lead to improvements through targeted exercises, even for unconventional body parts like feet

Does what does not kill you make you stronger?

59 implied HN points • 26 Jul 23
  1. The General Adaptation Syndrome proposed by Hans Selye has three stages: Alarm Reaction, Stage of Resistance, and Stage of Exhaustion.
  2. The fitness-fatigue model in the context of exercise suggests that the body has two competing systems: positive adaptation and exhaustion.
  3. Progressive Overload, a key concept in exercise science, involves introducing enough stimulus to induce adaptation and allowing rest periods to prevent injury and exhaustion.

A Causal Riddle

59 implied HN points • 17 Jul 23
  1. Lessons and diligent practice can cause language acquisition.
  2. It's not always clear if medication directly causes improvement in mental health.
  3. Understanding causality can be complex and lead to differing perspectives.