| Random Link ¯\_(ツ)_/¯ | ||
| Oct 4, 2021 | » | Journal Reviews on Fairness
8 min; updated Jun 7, 2026
MetaInstead of changing the data or learners in multiple ways and then see if fairness improves, postulate that the root causes of bias are the prior decisions that generated the training data. These affect (a) what data was selected, and (b) the labels assigned to the examples. They propose the \(\text{Fair-SMOTE}\) (Fair Synthetic Minority Over Sampling Technique) algorithm which (1) removes biased labels (via situation testing: if the model’s prediction for a data point changes once all of the data points' protected attributes are flipped, then that label is biased and the data point is discarded), and (2) rebalances internal distributions such that based on a protected attribute, examples are equal in both positive and negative classes. The method is just as effective in reducing bias as prior approaches, and its models achieve higher recall and F1 performance. Furthermore, \(\text{Fair-SMOTE}\) can simultaneously reduce bias for more than one protected attribute. ... |
| Nov 23, 2016 | » | What is Ergodicity?
3 min; updated Dec 25, 2024
A random process is ergodic if all of its statistics can be determined from a sample function of the process. That is, the ensemble averages equal the corresponding time averages with probability one. Role of Ergodicity in Human InferenceA newspaper has previously printed some inaccurate information, therefore, the newspaper is going to publish inaccurate information in the future. Fair; ensemble of published articles is more or less ergodic. More crimes are committed by black persons than by white persons, therefore each individual black person is not to be trusted? The ensemble of black people is not at all ergodic! ... |
| Dec 16, 2022 | » | Productivity for Software Engineers
8 min; updated Nov 16, 2024
Measuring ProductivityDevs’ Diverging Perceptions of ProductivityGot interested in measuring it two years into my SWE career. Initially rated myself based on % of completed daily objectives. The objectives didn’t necessarily correspond to work items. Stopped because the numbers were high even on days when I didn’t feel productive. Switched to “stuff that affects others” (later came to know these are “function points”). More intuitively captures non-code objectives, e.g., aligning folks on feature specs. Numbers seem low; never a double digit month. Open question on how to incorporate code reviews into my daily rating. ... |
| Jan 13, 2024 | » | March
11 min; updated Feb 4, 2024
Growing Up in Segregated AmericaThe thing is, when I was young, there wasn’t much of a civil rights movement. I wanted to work at something, but I grew up in rural Alabama. My parents knew it would be dangerous to make any waves. “Stay out of trouble. Don’t get in white people’s way.” ... |
| Mar 3, 2021 | » | LLMs: Stochastic Parrots 🦜 and How (Not) to Use Them
10 min; updated Dec 14, 2023
was written in a period when NLP practitioners are producing bigger (# of parameters; size of training data) language models (LMs), and pushing the top scores on benchmarks. The paper itself was controversial because it led to Gebru being fired from Google, following disagreements with her managers on conditions (withdraw, or remove Google-affiliated authors) for publishing the paper. A lot changed since mid-2021, when I initially wrote this page. OpenAI’s ChatGPT took the world by storm – reaching 123m MAU less than 3 months after launch and becoming the fastest-growing consumer application in history (TikTok took 9 months to hit 100m MAU). ... |
| May 2, 2020 | » | On Learning
11 min; updated May 27, 2023
Mental Attitude While LearningDistinguish Mere Facts From Conclusions or OpinionsDiscriminate between mere statements of facts, necessary conclusions which follow therefrom, and mere opinions which they seem to render reasonable. There’s no need to perform an experiment to verify that the atomic weight of oxygen is 16. That the sum of the angles of a plane triangle equals two right angles is not a mere fact, but an inevitable truth. ... |
| Jan 23, 2021 | » | Research on Privacy Enhancing Techniques
2 min; updated Feb 12, 2023
Journalsnote that prediction services can still make accurate predictions using a fraction of the data collected from a user device. They propose Cloak, which suppresses non-pertinent features (i.e. those features which can consistently tolerate addition of noise without degrading utility) to the prediction task. Cloak has a provable degree of privacy, and unlike cryptographic techniques, does not degrade prediction latency. Using the training data, labels, a pre-trained model and a privacy-utility knob, they (1) find the pertinent features through perturbation training, and (2) learn utility-preserving constant values for suppressing the non-pertinent data. During the inference phase, they efficiently compute a sifted representation which is sent to the service provider. ... |
| Jul 13, 2021 | » | On Becoming a Better Point Guard
10 min; updated Feb 12, 2023
Objective: Learn to read the game and make smart predictions. Reaction times are overrated; perception and prediction rule . Offensive Playmaking and Scoringtracks the following play types: transition, isolation, pick & roll ball handler, pick & roll roll man, post up, spot up, handoff, cut, off screen and put-backs. |
| Jan 11, 2022 | » | Journal Reviews
5 min; updated Feb 12, 2023
Link PredictionGiven network-structured data, predict whether a link exists between two nodes. Applications include: predicting drug-drug interactions (common in treating patients with complex/co-existing diseases) as they may cause changes in the drugs’ pharmacological activity . ... |
| Jan 15, 2022 | » | Software Engineering Journal Reviews
9 min; updated Feb 12, 2023
Formal Software DesignAlloy is an open-source language and analyzer for software modeling. An Alloy model is a collection of constraints that describe a set of structures, e.g. all possible security configurations of a web application. Alloy’s tool, the Alloy Analyzer is a solver that takes the constraints of a model and finds structures that satisfy them. The Alloy Analyzer leverages a SAT solver, and this precludes Alloy from analyzing optimization problems. propose AlloyMax, an extension of Alloy that can analyze problems with optimal solutions, soft constraints and priorities. AlloyMax adds new language constructs for specifying optimization problems, and uses an analysis engine that leverages a Maximum Satisfiability (MaxSAT) solver. also provide a translation mechanism from a first-order relational logic into a weighted conjunctive normal form (WCNF) that the MaxSAT solver expects. ... |
| Jan 20, 2023 | » | Gödel, Escher, Bach: An Eternal Golden Braid
10 min; updated Feb 12, 2023
Strange LoopsA Strange Loop occurs whenever, by moving upwards (or downwards) through the levels of some hierarchical system, we unexpectedly find ourselves right back where we started. The 3-voice Canon Per Tonos from Musikalisches Opfer BWV 1079 exhibits a strange loop. Successive modulations make one expect to hopelessly far from the starting key, but after six modulations, the original key of C minor is restored! |
| Dec 17, 2021 | » | 01. The Case for the Scout Mindset
6 min; updated Feb 12, 2023
Two Types of ThinkingSoldier MindsetReasoning is like defensive combat. Finding out you’re wrong means suffering a defeat. Seeks out evidence to fortify and defend your beliefs. Directionally motivated reasoning. When we want something to be true, we ask, “Can I believe this?” When we don’t want something to be true, we ask, “Must I believe this?” Scout MindsetReasoning is like mapmaking. Finding out you’re wrong means revising your map. Seeks out evidence that will make your map more accurate. ... |
| Apr 29, 2014 | » | The Gene-Free Model of Expertise
2 min; updated Sep 5, 2022
Reaction Times are OverratedElite athletes do not display superior reaction times to those of average people; most hover at 200ms. However, games are played at speeds where 200ms is too slow, e.g. 100-mph baseballs, 130-mph tennis serves, etc. Superior PerceptionAthletes outperformed novices when asked if there was a ball in a rapidly flashed slide. Some athletes could even tell the slide’s game! ... |
| Oct 12, 2021 | » | Reviews of WWW Proceedings
3 min; updated Sep 5, 2022
32nd ACM Conference on Hypertext and Social Media (HT ‘21)Reconstructing Ireland’s Lost History through the Beyond 2022 ProjectAims to create a virtual 3-D reconstruction of the “Record Treasury” of the Public Record Office of Ireland in Dublin. The original one was destroyed in Ireland’s Civil War of 1922. Reassembling will use copies, transcripts and records scattered among archival partners. |
| Dec 21, 2021 | » | We The Consumers
5 min; updated Sep 5, 2022
Product Differentiation and Price DiscriminationProduct differentiation seeks to distinguish a product from a competing product to make it more attractive to a specific target market. Price discrimination occurs when the same goods/services are sold at different prices from the same company. |
| Jan 1, 1976 | » | Toward a Theory of Medical Fallibility
5 min; updated Sep 5, 2022
Medical care is like the opposite of moving fast and breaking things. If it’s so taboo to admit error, then that could make errors more common because fewer people are learning from past errors. Norms for Scientific Activity and the Sources of Error“Science” is taken to mean “Natural Science”. Internal norms derive from a cognitive pursuit of science. They are:
External norms govern motives for participating in and using the results of scientific activity, e.g. ... |
| Sep 18, 2018 | » | On DNA Testing
3 min; updated Sep 5, 2022
The human genome has sequences of nucleotide base pairs that are repeated over and over again. At each locus of interest, a person has two sets of repeats inherited from each parent. Each possible difference at a locus is an allele. The combinations of the possible differences at multiple loci form a DNA profile that can be used to tie suspects to a crime scene. The Accuracy of DNA Testing is Wanting74/108 crime labs erroneously incriminated a suspect during a mock study. The reported match statistic for whether the match was coincidental varied over 100 trillion-fold. ... |
| Jan 27, 2020 | » | Tech and Democracy
5 min; updated Sep 5, 2022
Political AdsCambridge Analytica (CA) paid people to take in-app survey; mined FB profile data including friends’ data; crafted tailored sensitive ads to sway-able voters. Elections are about emotions, not facts. Data science and social media can help us make sense of and manipulate the chaos. |
| Nov 14, 2020 | » | Software Dependencies
6 min; updated Sep 5, 2022
Dependency ManagementGolang introduced a new library referencing mode to overcome limitations of the old one. While the two library modes are supported by Golang, they are incompatible, e.g. dependency management (DM) issues, reference inconsistencies, build failures, etc. did an empirical study that resulted in HERO, an automated technique to detect DM issues and suggest fixes. Applied to 19k Golang projects, HERO detected 98.5% on a DM issue benchmark, and found 2,422 new DM issues in 2,356 Golang projects. They reported 280 issues, and almost all of the fixes have adopted HERO’s fixing suggestions. ... |
| Oct 4, 2021 | » | Online Markets
4 min; updated Sep 5, 2022
WWW ‘21: The Web Conference 2021REST: Relational Event-Driven Stock Trend ForecastingREST, an event-driven stock trend forecasting framework, that overcomes two limitations of existing event-driven models. Models the stock context, and learns the effect of event information on the stocks under different contexts. Constructs a stock graph and designs a new propagation layer to propagate the effect of event information from related stocks. #stock-trend-forecasting #computational-finance The value of stock trend forecasting is not unanimous, e.g. Malkiel contends that forecasting is a fool’s game , while Simon’s RenTech is all about the math . But it seems like the question is an empirical one, and therefore, a good answer should exist. Why are there opposing camps? ... |
| Oct 28, 2021 | » | Socio-Economic Equity in Tech
7 min; updated Sep 5, 2022
When it comes to STEM diversity goals, Asian American and Pacific Islander (AAPI) tends to be excluded from the URMs under discussion, e.g. . However, AAPI as a blanket term obscures the struggles of member groups, e.g. \(62\%\) of AAPI adults aged 24 and older have an associate’s degree or higher, compared to 28% of Native Hawaiians and Pacific Islanders of the same age. AA comprises \(\approx 50\) ethnic groups, while PI has \(\approx 20\). ... |
| Dec 4, 2021 | » | As the Last I May Know
4 min; updated Sep 5, 2022
As the Last I May Know.
Shi Lian Huang.
Memorable PointsBackground: Sere missiles can wipe out a city completely. Nyma’s nation has been the only recipient of a seres strike. They’ve since acquired their own seres stockpile. They’re engaged in war with an state that lacks seres weapons. ... |
| Jul 2, 2022 | » | Debugging
7 min; updated Sep 5, 2022
Debugging 101Definition? Debugging involves inspecting a program’s internal state.
|
| Jan 2, 2022 | » | 04. Changing Your Mind
6 min; updated Sep 5, 2022
How to Be Wrongfound that experts were barely able to forecast better than random chance. However, a small subset of people (coined “superforecasters” ) were better. In a competition, they beat teams of top professors and CIA professional analysts. These superforecasters were not smarter than everyone else nor did they have more knowledge/experience, they were great at being wrong. Change your mind a little at a time. Seeing the world in shades of grey is less stressful, as the experience of encountering evidence against one of your beliefs is not high stakes. ... |
| Jan 1, 2022 | » | 03. Thriving Without Illusions
3 min; updated Sep 5, 2022
Coping with RealityThis chapter felt a bit vacuous with its use of anecdotes to advance points. Criticism of studies that argue that self-deception makes us happier: conflation of positive illusions and positive beliefs; unfounded definitions of what counts as self-deception; results that can be equally well-explained by something else. There are alternative coping strategies that don’t involve self-deception: making a hypothetical plan about some unpleasant and unavoidable thing; noticing silver linings but not to the point of sweet lemons; admitting that things could be worse. ... |
| Dec 28, 2021 | » | 02. Developing Self-Awareness
8 min; updated Sep 5, 2022
Signs of a ScoutThese Don’t Make You a ScoutFeeling objective. The more objective you feel, the more you trust your own intuitions and opinions as accurate representations of reality, and the less inclined you are to question them. Being smart and knowledgeable. It’s not a case of “if people were smarter and well-informed, they’d realize their errors”. For example, found that polarization (on political fronts) on anthropogenic climate change increases as scientific intelligence increases. found similar patterns: political & religious polarization on stem cell research, the Big Bang, human evolution; political polarization on climate change; little evidence of political or religious polarization on nanotechnology and genetically modified foods. ... |
Enlightening to see the struggles addressed by the civil rights movement through a child’s viewpoint. I can relate to John Lewis more at a young age because we have shared experiences, but his were in a segregated environment whereas mine weren’t.