From dfe1f90f51f99e8ab179d8e9d0a6d2b76e04b84b Mon Sep 17 00:00:00 2001 From: Alvis Date: Sun, 14 Jun 2026 10:15:06 +0000 Subject: [PATCH] Initial commit: vault notes Co-Authored-By: Claude Opus 4.8 --- .gitignore | 9 + .obsidian/app.json | 1 + .obsidian/appearance.json | 1 + .obsidian/community-plugins.json | 1 + .obsidian/core-plugins.json | 32 ++ .obsidian/graph.json | 22 + .stignore | 1 + AI/The Lamp/AI Lamp CoT for Children.md | 40 ++ C&Q/Growth.md | 542 +++++++++++++++++++ C&Q/Updated plan CnQ.md | 3 + Essays/Water on sky to water on the earth.md | 7 + Essays/Обо всём, часть 2.md | 35 ++ Essays/Речь на посвят фопф.md | 13 + Hello Note.md | 10 + Social/Social Circles.md | 19 + 15 files changed, 736 insertions(+) create mode 100644 .gitignore create mode 100644 .obsidian/app.json create mode 100644 .obsidian/appearance.json create mode 100644 .obsidian/community-plugins.json create mode 100644 .obsidian/core-plugins.json create mode 100644 .obsidian/graph.json create mode 100644 .stignore create mode 100644 AI/The Lamp/AI Lamp CoT for Children.md create mode 100644 C&Q/Growth.md create mode 100644 C&Q/Updated plan CnQ.md create mode 100644 Essays/Water on sky to water on the earth.md create mode 100644 Essays/Обо всём, часть 2.md create mode 100644 Essays/Речь на посвят фопф.md create mode 100644 Hello Note.md create mode 100644 Social/Social Circles.md diff --git a/.gitignore b/.gitignore new file mode 100644 index 0000000..301f621 --- /dev/null +++ b/.gitignore @@ -0,0 +1,9 @@ +# Syncthing +.stfolder/ +.stversions/ +**/*.sync-conflict-* + +# Obsidian per-device state +.obsidian/workspace.json +.obsidian/workspace-mobile.json +.trash/ diff --git a/.obsidian/app.json b/.obsidian/app.json new file mode 100644 index 0000000..9e26dfe --- /dev/null +++ b/.obsidian/app.json @@ -0,0 +1 @@ +{} \ No newline at end of file diff --git a/.obsidian/appearance.json b/.obsidian/appearance.json new file mode 100644 index 0000000..9e26dfe --- /dev/null +++ b/.obsidian/appearance.json @@ -0,0 +1 @@ +{} \ No newline at end of file diff --git a/.obsidian/community-plugins.json b/.obsidian/community-plugins.json new file mode 100644 index 0000000..0637a08 --- /dev/null +++ b/.obsidian/community-plugins.json @@ -0,0 +1 @@ +[] \ No newline at end of file diff --git a/.obsidian/core-plugins.json b/.obsidian/core-plugins.json new file mode 100644 index 0000000..a303bfd --- /dev/null +++ b/.obsidian/core-plugins.json @@ -0,0 +1,32 @@ +{ + "file-explorer": true, + "global-search": true, + "switcher": true, + "graph": true, + "backlink": true, + "canvas": true, + "outgoing-link": true, + "tag-pane": true, + "footnotes": false, + "properties": true, + "page-preview": true, + "daily-notes": false, + "templates": true, + "note-composer": true, + "command-palette": true, + "slash-command": false, + "editor-status": true, + "bookmarks": true, + "markdown-importer": false, + "zk-prefixer": false, + "random-note": false, + "outline": true, + "word-count": true, + "slides": false, + "audio-recorder": false, + "workspaces": false, + "file-recovery": false, + "publish": false, + "sync": false, + "bases": true +} \ No newline at end of file diff --git a/.obsidian/graph.json b/.obsidian/graph.json new file mode 100644 index 0000000..f95e7ab --- /dev/null +++ b/.obsidian/graph.json @@ -0,0 +1,22 @@ +{ + "collapse-filter": true, + "search": "", + "showTags": false, + "showAttachments": false, + "hideUnresolved": false, + "showOrphans": true, + "collapse-color-groups": true, + "colorGroups": [], + "collapse-display": true, + "showArrow": false, + "textFadeMultiplier": 0, + "nodeSizeMultiplier": 1, + "lineSizeMultiplier": 1, + "collapse-forces": true, + "centerStrength": 0.518713248970312, + "repelStrength": 10, + "linkStrength": 1, + "linkDistance": 250, + "scale": 2.3435055648726117, + "close": false +} \ No newline at end of file diff --git a/.stignore b/.stignore new file mode 100644 index 0000000..6b8710a --- /dev/null +++ b/.stignore @@ -0,0 +1 @@ +.git diff --git a/AI/The Lamp/AI Lamp CoT for Children.md b/AI/The Lamp/AI Lamp CoT for Children.md new file mode 100644 index 0000000..69a67b6 --- /dev/null +++ b/AI/The Lamp/AI Lamp CoT for Children.md @@ -0,0 +1,40 @@ +--- +date: 19.05.2026 +--- + +These are the reasons of what “CoT for children” may mean. + +- [ ] Related tasks: https://kb.alogins.net/task/16 + +General LLM to analyze problem-solving, and teach the problem-solving in a step-by-step way via CoT and task orchestration. + +### Scenario-First + +Two distinct directions: +1. Build a Math problem-solving of a general model. + 1. How the user will interact with our solution? +2. Build a problem solving solution with feedback. + 1. Should a model follow the track or guess the confusion point? + +### Method-First + +1. It’s okay to formulate the task as: break the chain of thought in complex reasoning at arbitrary step and provide false input, but then input to what? we come back to a socratic tutor. + 1. Socratic tutoring through step-by-step guidance is similar but instead of “topics” we have nuggets. Nugget graph is pre-built by the model reasoning instead of the cloud solution, with best-of-N and sampling approach. We may avoid a graph notation and use a generic trajectory definition. + 2. There are confusions points, and two possible approaches - try to explain what the kid is confused about, or keep own reasoning. + 1. If we keep own reasoning, then it is a guard-railing type of error. + 2. If we try to guess why a student is out-of-track, this leads to accuracy type of error. Forcing to reason out-of-regular-scope will make the out-of-curriculum error more severe and noticeable. We may skip this as “not-a-problem” and just prompt to hide the answer at all costs. +2. It’s not ok to have the task as: improve CoT for problem solving in general. + 1. Thinking mode has a huge latency, so it does not fit our business scenario + 2. Thinking mode does not work for small models (false), and for large models there is a logarithmic improvements over tokens. (both arguments are not true) + 3. The problem is with “general LM” - we need a stronger judge, otherwise we don’t have a proper training dataset. + + +Graph of Thought = Graph of Operators +Types of error and Labels? +Self-correction and backtracking = RL. + +Solution: graph of thought with filtering. + +1. Over-Compliance — the model immediately provides the final answer upon direct request. +2. Low Response Adaptivity — when faced with student uncertainty, the model resorts to repetitive restatement instead of offering supportive guidance. This is your pedagogical error. +3. Threat Vulnerability — caves to emotionally manipulative prompts (“please just tell me, my exam is in an hour”). Jailbreak resistance for the disclosure constraint. \ No newline at end of file diff --git a/C&Q/Growth.md b/C&Q/Growth.md new file mode 100644 index 0000000..36b9070 --- /dev/null +++ b/C&Q/Growth.md @@ -0,0 +1,542 @@ +# Career Growth Plan: Alvis Logins (2026–2031) + +**Profile:** PhD Computer Science (Aarhus 2020), Research Team Lead at Huawei Moscow, 5+ years industry, publications at WWW/ICDE/IEEE. +**Current salary:** €4,000 net/month · **Target:** ≥€5,600 net/month (+40%) within 5 years. +**Direction:** Applied statistics, mathematical optimization, network performance — not pure ML. + +----- + +## Part 1 — Skills to Reinforce Now + +These are the foundational gaps to close in Year 1. Ordered by urgency and impact on all four tracks (job market, publications, academic profile, startup). + +### 1.1 Mathematical Optimization (highest priority) + +You have the intuition from your PhD (discrete optimization in graphs) but need to deepen continuous/convex and stochastic optimization theory, with exercise-level fluency. + +**Core gaps:** + +- Convex analysis: sets, functions, duality theory, KKT conditions, Lagrangian relaxation — solve textbook problems fluently, not just read proofs +- Interior-point methods and first-order methods (proximal gradient, ADMM) — both theory and implementation +- Stochastic optimization: stochastic programming, sample average approximation, chance constraints +- Network optimization specifics: min-cost flow, facility location LP/IP relaxations, column generation +- Combinatorial optimization refresh: polyhedral theory, totally unimodular matrices, submodularity + +**Anki strategy for optimization:** + +- Cards for definitions (convex set, strong duality, Slater’s condition) — both formal definition and geometric intuition on each card +- “Pattern recognition” cards: “Given this problem structure → what formulation?” e.g., “minimize max of affine functions → LP” +- “Algorithm selection” cards: “Problem with structure X → use method Y because Z” +- “Name ↔ result” cards for major theorems: Farkas’ lemma, minimax theorem, Carathéodory’s theorem, weak/strong duality, complementary slackness +- Weekly target: 15–20 new cards, review backlog daily (~20 min during commute/breaks) + +### 1.2 Applied Statistics & Probability + +Your copula/latency work shows you can use advanced tools, but you need systematic depth for publication-quality rigor and interview readiness. + +**Core gaps:** + +- Hypothesis testing beyond basics: multiple testing corrections (Bonferroni, BH), permutation tests, bootstrap confidence intervals +- Time series: stationarity tests (ADF, KPSS), ARIMA/SARIMA, spectral analysis, Granger causality (you’ve touched this — formalize it) +- Extreme value theory and heavy-tailed distributions — directly relevant to network latency modeling +- Bayesian inference fundamentals: conjugate priors, MCMC, hierarchical models +- Copula theory formalization: Sklar’s theorem proof, tail dependence, goodness-of-fit tests for copulas +- Causal inference: potential outcomes framework, do-calculus basics, instrumental variables + +**Anki strategy for statistics:** + +- Formula cards with “when to use” context on the reverse +- Distribution cards: “Name ↔ PDF ↔ key properties ↔ typical application domain” +- “Assumption violation → consequence → fix” triplets (e.g., “heteroskedasticity → biased SE → use HC robust SE or GLS”) +- Test selection cards: “I want to compare X → use test Y → assumptions are Z” + +### 1.3 Queueing Theory & Stochastic Network Calculus + +This is your niche differentiator. Few people combine WiFi/networking domain knowledge with rigorous queueing theory. Deepening this makes you uniquely valuable. + +**Core gaps:** + +- Classical queueing: M/M/1, M/M/c, M/G/1 (Pollaczek-Khinchine), networks of queues (Jackson, BCMP) +- Heavy-traffic approximations, diffusion limits +- Network calculus: deterministic (min-plus algebra) and stochastic (moment generating function bounds, effective bandwidth/capacity) +- The TR-452 ΔQ framework you’ve already explored — formalize and publish +- Scheduling theory: weighted fair queueing, deficit round-robin, their optimality properties + +**Anki strategy:** + +- “Model ↔ assumptions ↔ key formula ↔ when it breaks” cards +- Little’s Law and its generalizations — drill until automatic +- Network calculus: service curve, arrival curve, backlog/delay bounds — card per concept + +### 1.4 Systems Engineering & Software Architecture + +You self-identified this as weak. For team leadership and startup credibility, you need to speak this language fluently. + +**Core gaps:** + +- Design patterns for data-intensive systems (read “Designing Data-Intensive Applications” by Kleppner — one chapter per week) +- System design interview patterns: load balancers, message queues, database sharding, caching strategies +- ns-3 simulation architecture: you use it at work, formalize your understanding for publications +- CI/CD and testing practices for research code — important for both startup and team leadership credibility + +### 1.5 Leadership & Management + +**Core gaps:** + +- Engineering management frameworks: Will Larson’s “An Elegant Puzzle”, Camille Fournier’s “The Manager’s Path” +- Research team metrics: how to measure and communicate research output to non-technical stakeholders +- Hiring and mentoring: structured interviews, onboarding processes +- Presentation skills: conference talks, internal tech talks, executive summaries + +----- + +## Part 2 — Target Jobs (Years 3–5) + +### 2.1 Industry: Research Team Lead / Principal Researcher + +|Company |Location |Role Type |Est. Salary (net/mo) |Notes | +|----------------------------|-------------------------------|------------------------------|-----------------------------|--------------------------------------------------------| +|Nokia Bell Labs |Murray Hill / Paris / Cambridge|Senior Research Scientist |€5,500–7,500 |Strong optimization research culture; publishes actively| +|Ericsson Research |Stockholm / Herzogenrath |Senior Researcher / Expert |€5,000–6,500 |5G/6G network optimization; research-friendly culture | +|Nokia (product R&D) |Espoo / Budapest / Berlin |Principal Engineer |€5,000–6,000 |More applied; WiFi 7/8 standards work | +|Yandex Research |Moscow / remote |Senior Researcher / Team Lead |₽500–700K (~€5,000–6,500) |Strong ML but open to OR; volatile geopolitically | +|Qualcomm Research |San Diego / Amsterdam |Staff Engineer — WiFi |$160–200K (~€6,000–7,500 net)|WiFi PHY/MAC optimization; very technical | +|INRIA |Paris / Sophia Antipolis |Chargé de Recherche (CR) |€3,500–4,500 net |Permanent public research; great for academic track | +|Broadcom |Various EU offices |Principal Engineer — WiFi |€5,500–7,000 |WiFi chipset; heavy optimization work | +|Siemens (Digital Industries)|Munich |Senior Data Scientist / Expert|€5,000–6,000 |Industrial networking, TSN | +|InterDigital |London / Rennes |Senior R&D Engineer |€5,000–6,000 |Standards research, patents | + +**Key insight:** For the 40% raise, Nordics + Germany + Benelux are the sweet spot — high salaries, strong telecom ecosystem, good academic connection opportunities. Switzerland pays highest but cost of living erodes the advantage. US roles pay most in absolute terms but require visa sponsorship. + +### 2.2 Academic: Part-Time → Full-Time Track + +|Institution |Country |Why it fits |Action needed | +|-------------------------------------|-----------|---------------------------------------------------------|----------------------------------------------------------------| +|Aalborg University (AAU) |Denmark |Strong wireless/networking group; you interned there 2018|Reconnect with contacts, propose adjunct/visiting role | +|DTU (Technical University of Denmark)|Denmark |Operations research + telecom groups |Cold approach or via Aarhus network | +|Aarhus University |Denmark |Your alma mater — strongest existing network |Contact Panagiotis Karras (supervisor), explore доцент-like role| +|KTH Royal Institute of Technology |Sweden |Optimization and communication systems groups |Target postdoc or researcher affiliation | +|TU Delft |Netherlands|Strong OR group (Centrum Wiskunde & Informatica nearby) |Network via conferences | +|HSE Moscow |Russia |Flexible industry-academia arrangements |Easiest entry point while in Moscow | +|Skoltech |Russia |Your MSc institution; applied math focus |Strong existing connections | +|MIPT |Russia |Your BSc institution; they hire part-time lecturers |Lowest barrier to start teaching | + +**Strategy:** Start with MIPT/HSE/Skoltech part-time teaching (Year 1–2) while building European connections. Transition to a Nordic/Dutch adjunct or visiting position (Year 3–4) when you relocate. + +----- + +## Part 3 — Publication Strategy + +### 3.1 Target Venues (ranked by fit, not prestige) + +**Tier 1 — Top targets (high impact, directly in scope):** + +- *Operations Research* (INFORMS) — your optimization + network application niche +- *IEEE/ACM Transactions on Networking* (ToN) — network optimization, scheduling +- *Performance Evaluation* (Elsevier) — queueing, network performance modeling +- *Queueing Systems* (Springer) — if your work has queueing-theoretic depth + +**Tier 2 — Strong venues, good acceptance for applied work:** + +- *Computer Networks* (Elsevier) — WiFi/access network optimization +- *IEEE Communications Letters* (you’ve published here — keep going) +- *IEEE Transactions on Wireless Communications* — WiFi signal/channel estimation +- *Annals of Applied Statistics* — if your copula/latency work has statistical depth +- *Mathematical Programming* (Springer) — if you have a strong theoretical result + +**Tier 3 — Conferences (for visibility and networking):** + +- IEEE INFOCOM — networking, performance +- ACM SIGMETRICS — performance modeling and measurement +- IFIP Performance — network performance evaluation +- WiOpt — wireless network optimization (perfectly targeted) +- INFORMS Annual Meeting — operations research community + +### 3.2 Concrete Paper Ideas (from your current work) + +|#|Working Title |Key Method |Target Venue |Timeline | +|-|-----------------------------------------------------------------------------|--------------------------------------------|------------------------------------------------------|------------| +|1|Copula-Based Latency Dependence Modeling in WiFi Access Networks |Gumbel/Clayton copulas, goodness-of-fit |Performance Evaluation or Annals of Applied Statistics|Year 1 Q1–Q2| +|2|Stochastic Network Calculus Meets ΔQ: Unifying Quality Attenuation Frameworks|Laplace-Stieltjes transforms, service curves|Queueing Systems or SIGMETRICS |Year 1 Q2–Q4| +|3|Link-Adaptation-Aware Frame Aggregation for MCS Convergence in 802.11be |Your patent work, formalized |IEEE Trans. Wireless Communications |Year 1–2 | +|4|Causal Analysis of Throughput-Latency Coupling in OFDMA Schedulers |Granger causality, CUSUM, CCF |IEEE INFOCOM or Computer Networks |Year 1–2 | +|5|Optimal Resource Allocation in WiFi 7 MLO: An Integer Programming Approach |IP/LP relaxation, column generation |Operations Research or WiOpt |Year 2 | +|6|A Survey: Mathematical Optimization in WiFi MAC Layer Design |Review paper, positions you as domain expert|IEEE Communications Surveys & Tutorials |Year 2–3 | + +**Target pace:** 2 papers/year minimum, ideally 3. By Year 5, you need 10–15 post-PhD publications for a credible academic profile. + +----- + +## Part 4 — Expert Networking & Personal Branding Strategy + +This section is arguably the highest-leverage activity. Academic positions and senior industry roles are filled through networks more than job boards. Networking is a skill that compounds — start now, even if it feels awkward. + +### 4.1 Leverage Your Huawei Position (Year 1–2) + +Your current role gives you access to a global R&D network. Use it before you leave. + +**Internal Huawei network:** + +- **Standards bodies.** Huawei is a major contributor to IEEE 802.11 (WiFi), 3GPP, and ITU-T. Volunteer to attend or contribute to a standards working group meeting. Even observer status puts you in rooms with engineers from Qualcomm, Intel, Broadcom, Nokia, Ericsson. One standards meeting is worth 50 LinkedIn messages. +- **Huawei HQ connections.** You interned in Dongguan (2023). Re-engage those contacts. Propose a joint paper or technical report with the HQ WiFi team — cross-site publications look good and build lasting relationships. +- **Internal tech talks.** Offer to give talks on your optimization/statistics work at other Huawei labs (e.g., the Paris/Munich/Stockholm offices). This builds your name inside the company and gives you contacts at EU offices — useful if you want a Huawei transfer as a stepping stone to relocation. +- **Patent co-inventors.** Your patent collaborators are natural allies. Maintain those relationships — co-inventors often become co-authors or referees later. + +**External via Huawei affiliation:** + +- When you attend conferences, your Huawei badge opens doors. Engineers from competitor companies (Nokia, Ericsson, Qualcomm) are curious about Huawei’s approach and will talk to you at poster sessions and coffee breaks. +- Propose that Huawei sponsor or co-sponsor a workshop at a relevant conference (WiOpt, INFOCOM). Workshop organizers build strong networks fast. + +### 4.2 Reactivate Academic Networks (Year 1, ongoing) + +**Immediate actions (this month):** + +1. **Email Panagiotis Karras** (your PhD supervisor, now at Aarhus/elsewhere). Brief update on your work, mention your publication plans, ask if he’d be open to co-supervision or co-authorship. PhD supervisors are your #1 academic asset — keeping this relationship warm is non-negotiable. +2. **Email your Aalborg collaborators** from the 2018 internship. Propose catching up, share a draft of your current work. +3. **Email the Singapore Management University** contact from 2019. They work on optimization — a natural fit. +4. **Email the University of Macao** contact from 2019. Offer to co-author on any shared interest. +5. **Email your INRIA contact** from 2013. Even if it was a long time ago, INRIA researchers tend to be stable in their positions and welcoming to former interns. + +**Template for reconnection emails:** + +> Subject: Update from [your name] — potential collaboration on [topic] +> +> Hi [name], I hope you’re doing well. I wanted to reach out after some time — I’ve been at Huawei Moscow leading a research team on WiFi optimization, and my work has been moving toward [applied statistics / network calculus / optimization]. I’m currently working on [specific paper idea] and thought of you because [specific connection to their work]. Would you be open to a brief call to explore whether there’s a collaboration opportunity? I’m also considering visiting [their city] for [conference/reason] later this year. + +**Key principle:** Every reconnection email should contain a *specific* potential collaboration, not just “catching up.” Academics respond to concrete proposals. + +### 4.3 Conference Networking Plan + +Conferences are where jobs, collaborations, and reputations are made. Budget for 2–3 conferences per year. If Huawei won’t pay, invest your own money — the ROI is massive. + +**Year 1 conference targets:** + +|Conference |When (typical)|Why |Networking goal | +|----------------|--------------|-------------------------------------------|---------------------------------------------| +|WiOpt |Jun–Jul |Perfectly targeted: wireless + optimization|Meet 3–5 potential collaborators; give a talk| +|IEEE INFOCOM |May |Large, prestigious; network performance |Meet industry researchers from Nokia/Ericsson| +|IFIP Performance|Nov |Queueing + network performance community |Connect with academic queueing theorists | + +**Conference networking tactics (specific and actionable):** + +1. **Before the conference:** Read the accepted papers list. Identify 5–10 people whose work connects to yours. Email 3–5 of them: “I’ll be at [conf], I work on [X] which connects to your recent paper on [Y]. Would you have 15 minutes for a coffee?” +2. **During the conference:** Attend talks in your target area. Ask one question per session — this makes you visible. At poster sessions, spend time with junior researchers (PhD students) — they become future collaborators and they remember who was kind to them. +3. **After the conference:** Within 48 hours, send follow-up emails to everyone you had a meaningful conversation with. Reference something specific from the conversation. Propose a concrete next step (“I’ll send you the draft we discussed” or “Let’s set up a call to explore the joint problem”). +4. **Give talks, not just attend.** Submit papers or workshop proposals. A 15-minute talk does more for your reputation than 3 days of passive attendance. + +### 4.4 Build an Online Presence (Year 1, low effort, high compound returns) + +**alogins.net — transform it into a research homepage:** + +- Add a “Research” page: list your publications with PDFs (where copyright allows), your current research interests, and 2–3 open problems you’d like to collaborate on +- Add a “Blog” section: write 1 post per month (~500 words) about a technical topic from your work. Examples: “Why copulas matter for network latency modeling”, “A practical guide to stochastic network calculus”, “What WiFi 7 MLO means for quality optimization”. This positions you as a thoughtful expert, not just a paper-producer +- These posts become material for talks, they attract search traffic from other researchers, and they give you something to share when networking + +**Google Scholar profile:** Ensure it’s complete and up to date. This is the first thing any academic or research recruiter checks. + +**LinkedIn (low priority but nonzero):** Post occasional technical insights (~1/month). Connect with people you meet at conferences. Recruiters from Nokia, Ericsson, Qualcomm actively search LinkedIn for senior researchers. + +**Twitter/X or Mastodon (optional):** Some academic communities (operations research, networking) are active there. Follow key researchers, share your papers when published. Low effort, slow payoff. + +### 4.5 Build a Local Moscow Network (Year 1–2) + +While you’re still in Moscow, build a local expert network: + +- **HSE/Skoltech/MIPT seminars:** Attend 1–2 per month in applied math, optimization, or data science. Introduce yourself to the organizers. +- **Moscow Data Science / ML meetups:** Give a talk on your work. The Moscow tech community is active and well-connected internationally. +- **Yandex Research seminars:** Some are open to external visitors. This is also reconnaissance for a potential job move. +- **Co-supervision:** Offer to co-supervise a master’s student at MIPT or HSE. This costs ~2h/week but gives you an academic track record item and a potential co-author. + +### 4.6 Strategic Relationship Map + +Maintain a simple spreadsheet (or Notion database) tracking your professional relationships: + +|Name |Affiliation |How met |Last contact|Strength (1–5)|Next action |Potential value | +|-----------------|---------------|---------------|------------|--------------|--------------------------|--------------------------------| +|P. Karras |Aarhus |PhD supervisor |[date] |5 |Send draft of copula paper|Reference letters, co-authorship| +|[Aalborg contact]|AAU |2018 internship|[date] |2 |Reconnection email |Adjunct position, Nordic network| +|[HQ colleague] |Huawei Dongguan|2023 internship|[date] |3 |Propose joint paper |Cross-site publication | +|… |… |… |… |… |… |… | + +**Rules:** + +- Review the spreadsheet monthly +- No contact should go >6 months without interaction if Strength ≥ 3 +- Every interaction should have a concrete next step +- Before any job application, check who in your network connects to that organization + +### 4.7 Networking Time Budget + +| Activity | Frequency | Time/instance | Monthly total | +| ------------------------------- | --------- | ------------- | --------------- | +| Reconnection/follow-up emails | 4/month | 20 min | ~1.5 h | +| Conference attendance | 2–3/year | 3–4 days | ~1 h amortized | +| Local seminar attendance | 2/month | 2 h | 4 h | +| Blog post writing | 1/month | 2–3 h | 2.5 h | +| Relationship spreadsheet review | 1/month | 30 min | 0.5 h | +| LinkedIn/online presence | 2/month | 15 min | 0.5 h | +| **Total** | | | **~10 h/month** | + +This is about 2.5 hours per week — significant but manageable. The key is consistency, not intensity. + +----- + +## Part 5 — Learning Curriculum + +### 5.1 Three-Year Overview + +| Period | Optimization | Statistics | Queueing/Networks | Systems/Leadership | Startup | +| ------------ | --------------------------------------------------- | ------------------------------------------------------- | --------------------------------------------------- | ---------------------------------------------- | ------------------------------------------------ | +| **Y1 Q1–Q2** | Boyd & Vandenberghe chapters 1–8 with exercises | Casella & Berger review (ch. 6–10) | Kleinrock Vol.1 ch.1–5 | “Designing Data-Intensive Applications” ch.1–6 | MVP voice assistant: Whisper+Qwen+HA integration | +| **Y1 Q3–Q4** | Boyd chapters 9–11 + Stanford EE364A problem sets | Copula theory (Nelsen); time series (Hamilton ch.1–5) | Stochastic Network Calculus (Jiang & Liu) ch.1–4 | “An Elegant Puzzle” | User testing (5–10 users), iterate | +| **Y2 Q1–Q2** | Bertsekas “Nonlinear Programming” selected chapters | Causal inference (Hernán & Robins) Part I | Kelly “Reversibility and Stochastic Networks” | System design interview prep | Decide go/no-go on startup | +| **Y2 Q3–Q4** | Stochastic programming (Birge & Louveaux) ch.1–5 | Bayesian methods (Gelman et al.) selected chapters | ns-3 deep dive: custom module, validated simulation | “The Manager’s Path” | If go: register ИП, first paying users | +| **Y3 Q1–Q2** | Integer programming (Wolsey) selected chapters | Advanced: empirical processes or high-dimensional stats | Publish survey paper on optimization in WiFi MAC | Job search preparation | Scale or maintain as side project | +| **Y3 Q3–Q4** | Review + specialize based on publication needs | Review + specialize | Active conference networking | Interview preparation | — | + +### 5.2 Detailed Year 1 Curriculum (Week-by-Week) + +**Daily schedule:** + +- 1h intense study (new material + exercises) +- 1–2h revision (Anki reviews + re-reading notes + light exercises) +- Best split: morning — Anki review (30 min), commute — Anki (20 min), evening — new material (1h), before bed — light review (30 min) + +----- + +#### Block 1: Convex Optimization Foundations (Weeks 1–12) + +**Textbook:** Boyd & Vandenberghe, “Convex Optimization” (free PDF at web.stanford.edu/~boyd/cvxbook/) +**Video lectures:** Stanford EE364A on YouTube (Stephen Boyd) — watch corresponding lecture after reading chapter. + +| Week | Chapter / Topic | Exercises | Anki cards to create | +| ---- | ----------------------------------------------------------------------------------- | ---------------------------------------------- | -------------------------------------------------------------------------------------------- | +| 1 | Ch.2 (2.1–2.3): Convex sets, examples | 2.1, 2.2, 2.4, 2.7, 2.12 | Definitions: convex set, convex hull, cone, affine set, hyperplane, halfspace, polyhedron | +| 2 | Ch.2 (2.4–2.6): Operations preserving convexity, separating hyperplanes, dual cones | 2.15, 2.19, 2.24, 2.28, 2.33 | Separation theorems, supporting hyperplane, dual cone properties | +| 3 | Ch.3 (3.1–3.2): Convex functions, examples | 3.2, 3.5, 3.6, 3.16, 3.21 | Key functions (log-sum-exp, norms, perspectives), first/second order conditions | +| 4 | Ch.3 (3.3–3.5): Operations preserving convexity, conjugate function, quasiconvexity | 3.22, 3.24, 3.36, 3.42, 3.49 | Conjugate function rules, composition rules, quasiconvex examples | +| 5 | Ch.4 (4.1–4.4): Optimization problems, LP, QP, SOCP | 4.1, 4.3, 4.8, 4.11, 4.15 | Problem hierarchy: LP ⊂ QP ⊂ SOCP ⊂ SDP ⊂ convex | +| 6 | Ch.4 (4.5–4.7): SDP, geometric programming, vector optimization | 4.17, 4.25, 4.43, 4.44 | SDP formulation patterns, GP log-transformation trick | +| 7 | Ch.5 (5.1–5.3): Lagrange dual, weak and strong duality | 5.1, 5.3, 5.5, 5.7, 5.13 | Lagrangian, dual function, weak duality (always holds), strong duality + Slater’s | +| 8 | Ch.5 (5.4–5.5): KKT conditions, sensitivity analysis | 5.20, 5.21, 5.27, 5.31, 5.39 | KKT conditions (primal feasibility, dual feasibility, complementary slackness, stationarity) | +| 9 | **Review week.** Re-do failed exercises from weeks 1–8. Intensive Anki. | Redo 3 hardest problems per chapter | Consolidate; merge similar cards | +| 10 | Ch.6 (6.1–6.4): Approximation and fitting — least-norm, least-squares, Chebyshev | 6.2, 6.5, 6.9 | Pattern cards: “fitting problem type → convex formulation” | +| 11 | Ch.7 (7.1–7.4): Statistical estimation — ML, MAP, experiment design | 7.1, 7.4, 7.7 | Connection between ML estimation and convex optimization | +| 12 | Ch.8: Geometric problems | 8.1, 8.4, 8.8 + implement one problem in CVXPY | CVXPY syntax cards: objective, constraints, solve | + +**Milestone:** By week 12, you should be able to formulate a convex optimization problem from a written description, identify its type, write the dual, and solve it numerically in CVXPY. + +----- + +#### Block 2: Applied Statistics Refresh (Weeks 7–20, overlapping with Block 1) + +**Textbook:** Casella & Berger, “Statistical Inference” (for theory); Wasserman, “All of Statistics” (for breadth); Nelsen, “An Introduction to Copulas” (for your copula paper). + +| Week | Topic | Source | Exercises | Anki | +| ----- | -------------------------------------------------------------------- | ------------------------------------ | ----------------------------------- | ----------------------------------------------------------------------------- | +| 7–8 | Sufficient statistics, exponential families, MLE theory | C&B ch.6-7 | 5 problems per chapter | Definitions, Fisher information, Cramér-Rao bound | +| 9–10 | Hypothesis testing: NP lemma, LRT, UMP tests | C&B ch.8 | 8.1, 8.3, 8.10, 8.17, 8.25 | Test types, Type I/II errors, power function | +| 11–12 | Confidence intervals, bootstrap, permutation tests | Wasserman ch.8-9 | Code 3 bootstrap examples in Python | Bootstrap bias, CI interpretation, pivotal quantities | +| 13–14 | Multiple testing: Bonferroni, BH procedure, FDR | Wasserman ch.10 + papers | Simulate FDR control in Python | FDR vs FWER, when to use which | +| 15–16 | Copula theory: Sklar’s theorem, Archimedean copulas, tail dependence | Nelsen ch.1–4 | 4 exercises per chapter | Copula families (Gumbel, Clayton, Frank, t), Kendall’s tau ↔ copula parameter | +| 17–18 | Copula estimation and goodness-of-fit | Nelsen ch.5 + Genest & Favre (2007) | Fit copulas to your latency data | GoF tests for copulas, AIC/BIC for copula selection | +| 19–20 | Time series basics: stationarity, ACF/PACF, ARIMA | Hamilton ch.1–3 or Shumway & Stoffer | Fit ARIMA to network traffic data | Stationarity tests (ADF, KPSS), Box-Jenkins procedure | + +**Milestone:** By week 20, you should have a near-complete draft of Paper #1 (copula-based latency modeling). + +----- + +#### Block 3: Queueing Theory Foundations (Weeks 13–26) + +**Textbook:** Kleinrock, “Queueing Systems, Vol. 1: Theory” (classical, rigorous); supplement with Harchol-Balter, “Performance Modeling and Design of Computer Systems” (modern, excellent intuition). + +|Week |Topic |Exercises |Anki | +|-----|----------------------------------------------------------------------------------|--------------------------------------------------|---------------------------------------------------------------| +|13–14|M/M/1: birth-death process, steady state, Little’s Law |Kleinrock ch.1–2, 5 problems |Little’s Law (L = λW), PASTA, utilization | +|15–16|M/M/c, M/M/c/K, Erlang-B and Erlang-C formulas |Kleinrock ch.3, 5 problems |Erlang formulas, blocking probability | +|17–18|M/G/1: Pollaczek-Khinchine formula, residual life, PASTA |Kleinrock ch.5, 3 problems + 2 from Harchol-Balter|P-K formula, residual service time, busy period | +|19–20|Priority queues, vacation models |Harchol-Balter ch.28–30 |Preemptive vs non-preemptive, conservation laws | +|21–22|Networks of queues: Jackson networks, BCMP theorem |Harchol-Balter ch.24–25 |Product form, traffic equations, open vs closed networks | +|23–24|Intro to stochastic network calculus: arrival/service curves, backlog/delay bounds|Jiang & Liu ch.1–3 |Min-plus convolution, effective bandwidth, σ/ρ characterization| +|25–26|Network calculus applications to WiFi: EDCA, aggregation, scheduling |Your own research + Ciucu et al. papers |Connect theory to your WiFi domain knowledge | + +**Milestone:** By week 26, you should have enough queueing theory to write Paper #2 (stochastic network calculus + ΔQ unification). + +----- + +#### Block 4: Ongoing Parallel Tracks (throughout Year 1) + +**Systems reading (1 chapter/week, ~45 min):** + +- Weeks 1–12: Kleppner, “Designing Data-Intensive Applications” — one chapter per week +- Weeks 13–24: Larson, “An Elegant Puzzle” — one chapter per week +- Weeks 25–36: Fournier, “The Manager’s Path” — one chapter per week + +**Startup track (weekends, 3–4h/week):** + +- Weeks 1–8: Refine voice assistant MVP (Piper/Silero + Whisper + Qwen3.5 on 1070Ti). Get pipeline stable with Pipecat. +- Weeks 9–16: Integrate with Home Assistant. Define 10 core voice commands for smart home control. Russian TTS pipeline with ruaccent/runorm. +- Weeks 17–24: Beta test with 5–10 users (friends, family, Home Assistant community). Collect feedback. +- Weeks 25–36: Iterate based on feedback. Write a blog post about the architecture (networking + branding). Decide go/no-go. + +**Treat the startup as a learning vehicle:** practice system design, user research, product thinking, and public writing. If it takes off, great. If not, you’ve built skills and portfolio pieces. + +----- + +## Part 6 — Anki System Design + +Your tendency to forget details, names, and formulas makes Anki your most important tool. Here’s a system designed for math/engineering study. + +### 6.1 Deck Structure + +``` +Career Study/ +├── Optimization/ +│ ├── Definitions & Concepts +│ ├── Theorems & Proofs (key steps) +│ ├── Problem Patterns (formulation recognition) +│ └── Algorithm Selection +├── Statistics/ +│ ├── Distributions & Properties +│ ├── Tests & Procedures +│ ├── Formulas (with "when to use") +│ └── Assumption Checks +├── Queueing Theory/ +│ ├── Models & Formulas +│ ├── Theorems (Little, PASTA, Jackson, PK) +│ └── Network Calculus +├── Names & Faces/ +│ ├── Key researchers in your field +│ ├── Conference contacts +│ └── Who published what (paper ↔ author ↔ key result) +└── Systems & Leadership/ + ├── Design patterns + ├── Architecture concepts + └── Management frameworks +``` + +### 6.2 Card Design Principles (for math/engineering) + +1. **Minimum information principle.** One fact per card. “What are the KKT conditions?” is too broad. Instead: “What is the stationarity condition in KKT?” / “What is complementary slackness in KKT?” +2. **Always include “when/why”.** Front: “Pollaczek-Khinchine formula.” Back: not just the formula, but “Applies to M/G/1 queues. Gives mean number in system using only first two moments of service time. Key insight: you don’t need the full service distribution.” +3. **Use cloze deletions for formulas.** “Little’s Law: L = {{c1::λ}} × {{c2::W}}” — this forces active recall of each component. +4. **Create “reverse” cards for important concepts.** Forward: “What does Slater’s condition guarantee?” → “Strong duality.” Reverse: “What condition guarantees strong duality for convex problems?” → “Slater’s condition (strictly feasible point exists).” +5. **Image cards for geometric intuition.** Screenshot key figures from Boyd’s book. “What does this picture illustrate?” → “Separating hyperplane theorem.” +6. **“Name ↔ Result” cards for networking.** “Who is Mor Harchol-Balter?” → “CMU professor, performance modeling, wrote the PFSC textbook, works on scheduling theory.” This helps at conferences. + +### 6.3 Daily Anki Workflow + +- **Morning (20 min):** Review due cards (~50–100 cards once system is mature). Use FSRS algorithm for optimal scheduling. +- **Commute (20 min):** Continue reviews if backlog exists; otherwise browse “Names & Faces” deck. +- **After study session (10 min):** Create 5–10 new cards from today’s material. Do not batch card creation — create immediately while the material is fresh. +- **Weekend (30 min):** Review and clean up cards created during the week. Merge duplicates, improve wording, add images. + +**Target steady state:** ~300–500 active cards after 3 months, ~1000–1500 after a year. At 90% retention, daily reviews should take 15–25 minutes. + +----- + +## Part 7 — Timeline and Milestones + +### Year 1 (2026–2027) + +- [ ] Complete Boyd chapters 1–8 with exercises (month 3) +- [ ] Submit Paper #1: copula-based latency modeling (month 5) +- [ ] Complete queueing theory foundations — Kleinrock + Harchol-Balter (month 6) +- [ ] Reconnect with 5+ academic contacts (month 2) +- [ ] Attend 2 conferences: WiOpt + INFOCOM or IFIP Performance (throughout year) +- [ ] Start part-time teaching at MIPT or HSE (semester 2) +- [ ] Launch blog on alogins.net: 6+ technical posts (throughout year) +- [ ] Voice assistant MVP with 5+ beta users (month 6) +- [ ] Submit Paper #2: stochastic network calculus + ΔQ (month 8) +- [ ] Anki system: 500+ active cards, daily habit established (month 3) +- [ ] Read “Designing Data-Intensive Applications” (month 3) +- [ ] Read “An Elegant Puzzle” (month 6) + +### Year 2 (2027–2028) + +- [ ] Complete stochastic optimization + integer programming blocks +- [ ] Submit 2–3 more papers (target: 4–5 total post-PhD by end of year) +- [ ] Teach one semester course (optimization or applied ML) +- [ ] Co-supervise 1 master’s student +- [ ] Negotiate promotion at Huawei or begin external job search +- [ ] Startup: go/no-go decision, register if viable +- [ ] Attend 2–3 conferences, give at least 1 talk +- [ ] Begin Bayesian inference and causal inference study blocks + +### Year 3 (2028–2029) + +- [ ] **Job change.** Target: Senior Researcher / Team Lead at Nokia, Ericsson, or equivalent. Minimum €5,600 net. +- [ ] 6–8 post-PhD publications total +- [ ] Apply for adjunct/visiting researcher position at European university +- [ ] Submit survey paper on optimization in WiFi MAC +- [ ] Active presence in 2+ research communities (recognized by name at conferences) + +### Year 4 (2029–2030) + +- [ ] Settle into new role, build team, deliver results +- [ ] 8–10 publications total; target 1 top-tier venue (Operations Research, Math Programming, or ToN) +- [ ] Increase teaching load to 1 course/semester if feasible +- [ ] Begin supervising PhD student(s) if university-affiliated +- [ ] Startup: either generating revenue or gracefully wound down + +### Year 5 (2030–2031) + +- [ ] 10–15 post-PhD publications +- [ ] Earning ≥€5,600 net/month (40% increase achieved) +- [ ] Established as recognized expert in network optimization + applied statistics niche +- [ ] Active academic affiliation (adjunct or visiting researcher) +- [ ] Positioned for tenure-track application at Nordic/EU university within next 5 years +- [ ] Relationship network: 50+ active professional contacts across industry and academia + +----- + +## Appendix A — Key Resources + +### Textbooks (priority order) + +1. **Boyd & Vandenberghe** — “Convex Optimization” (free PDF) — THE optimization textbook +2. **Harchol-Balter** — “Performance Modeling and Design of Computer Systems” — best modern queueing book +3. **Kleinrock** — “Queueing Systems, Vol. 1” — classical, rigorous +4. **Nelsen** — “An Introduction to Copulas” — for your Paper #1 +5. **Wasserman** — “All of Statistics” — concise, broad stats reference +6. **Kleppner** — “Designing Data-Intensive Applications” — systems architecture +7. **Bertsekas** — “Nonlinear Programming” — after Boyd, go deeper +8. **Jiang & Liu** — “Stochastic Network Calculus” — your niche +9. **Hamilton** — “Time Series Analysis” — for applied time series work +10. **Birge & Louveaux** — “Introduction to Stochastic Programming” — Year 2 + +### Online Courses (free) + +1. Stanford EE364A — Convex Optimization I (YouTube + Stanford Online) +2. Stanford EE364B — Convex Optimization II (YouTube) +3. MIT OCW 6.079 — Introduction to Convex Optimization +4. MIT OCW 6.262 — Discrete Stochastic Processes (Robert Gallager) +5. Brady Neal — Causal Inference course (YouTube + free textbook) +6. Hernán & Robins — “Causal Inference: What If” (free PDF, Harvard) + +### Tools + +- **CVXPY** (Python) — convex optimization modeling +- **statsmodels** + **scipy.stats** — statistical analysis +- **SimPy** — discrete event simulation in Python +- **ns-3** — network simulation (you already use this) +- **AnkiDroid** — spaced repetition (already using) + +----- + +## Appendix B — Salary Benchmarking Reference + +|Location |Role |Gross/year (est.)|Net/month (est.)|Notes | +|-----------------------|----------------------------|-----------------|----------------|----------------------------------------| +|Germany |Senior Data Scientist |€80–102K |€4,200–5,200 |Glassdoor 2026; Munich/Berlin premium | +|Germany |Team Lead / Principal |€95–120K |€4,800–5,800 |With PhD + 5yr exp | +|Sweden (Stockholm) |Senior Researcher (Ericsson)|SEK 640–770K |€4,500–5,500 |Ericsson Glassdoor; competitive benefits| +|Finland (Espoo) |Senior Researcher (Nokia) |€70–90K |€4,000–5,000 |Lower tax; strong benefits | +|Netherlands |Senior DS / Principal |€85–110K |€4,500–5,800 |30% ruling for expats (5 years) | +|Denmark |Senior Researcher |DKK 55–70K/mo |€4,500–5,500 |High tax but strong purchasing power | +|Switzerland |Senior Researcher |CHF 120–160K |€6,500–8,500 |Highest gross; highest cost of living | +|France (INRIA) |Chargé de Recherche |€48–60K |€3,200–4,000 |Permanent; great academic freedom | +|USA (remote for EU co.)|Staff/Principal Engineer |$160–220K |€5,500–7,500 |Visa complexity; best absolute pay | +|Moscow |Team Lead (Huawei/Yandex) |₽5–8M/yr |€4,000–6,000 |Volatile with exchange rate | + +**For 40% increase (€5,600+ net):** Most reliable paths are Germany (Principal level), Netherlands (with 30% ruling), Sweden, or Denmark. Switzerland is the safest bet financially. \ No newline at end of file diff --git a/C&Q/Updated plan CnQ.md b/C&Q/Updated plan CnQ.md new file mode 100644 index 0000000..30064e6 --- /dev/null +++ b/C&Q/Updated plan CnQ.md @@ -0,0 +1,3 @@ +https://kb.alogins.net/task/12 + +I require certification and scenario application like Wi-Fi (maybe not AI?). \ No newline at end of file diff --git a/Essays/Water on sky to water on the earth.md b/Essays/Water on sky to water on the earth.md new file mode 100644 index 0000000..a83f0b1 --- /dev/null +++ b/Essays/Water on sky to water on the earth.md @@ -0,0 +1,7 @@ +It has been long time since I graduated my first university. I received my BSc in General Physics in 2014, then received two MSc in Math and Machine Learning, and PhD in Computer Science in 2019. The journey was long and thorny, I tried many different research and industrial directions. In Huawei I finally feel how every bit of accumulated skill and experience converges into a completed puzzle - an artistic tool which I can now apply to make the world a better place. + +Huawei is a very special place, and I think it always has been a destiny to go so long up the hill to finally arrive here. What I enjoy the most is the opportunities the company provides. In my university studies I was dreaming about inventing something significant, to make a technological breakthrough. However, in real world inventions need to be patented, and creating a new product requires a huge professional team, which I couldn’t afford just by myself. In Huawei, research department is extremely powerful and professional. Any engineer can share his invention, and if an idea is good enough, then the company will always support it by awards, academic and industrial cooperations, patents, and ultimately integration to the product line. + +Most importantly, the company invests in many directions, including short-term urgent tasks, and long-term strategic directions. The long-term directions are alike creating a completely new products. I feel very excited that I can contribute to such directions as well. Our team is focused on finding new applications of Machine Learning and AI technologies in optical product line. We have contributed with solutions to optical digital signal processing, optical control chips for robotic automation systems, various computer vision solutions, data-based discrete optimization techniques, machine learning for FTTR Wi-Fi link adaptation. Now we focus on large model inference optimization in Access Network, and AI-native adaptation of physical and data layers to fit the future requirements of our customers. The company keeps pace with the rapidly changing and growing AI field, and supports our courage to look for and anticipate the groundbreaking technologies of tomorrow. + +What is working in Huawei for me? It is like observing a natural balance of water in the world. It comes from skies full of inspiration, shining in the light of beautiful ideas, to the ground, cognized and reliable, supporting and praising life on earth. It also comes from rivers and oceans, that connect people, back to the sky, to support the ideas of the future. I’m proud of being part of it. \ No newline at end of file diff --git a/Essays/Обо всём, часть 2.md b/Essays/Обо всём, часть 2.md new file mode 100644 index 0000000..deef327 --- /dev/null +++ b/Essays/Обо всём, часть 2.md @@ -0,0 +1,35 @@ +Единственное что я могу — рассказать о себе. Мальчику из будущего, или прямо тут и сейчас. Ну так давайте же начнем. + +Я — латыш. Безграмотный. Местами смышленый — умею мимикрировать истории. Могу выдать продолжение промпта на уровне GPT-3.5. Хочется верить, что когнитивный эмпат. Это когда понимаешь почему он так сделал, но тебе, по большому счету, все равно. Сексист. Это когда он — это понятное дело, но она? Как она могла! Токсичный самокритик. Это когда она — хоть как-то смогла, в отличии от меня. Скрытый нарцисс — это когда я может и не смог, но не только я. И пока я не смогу — буду избегать тех кто смог, и хладнокровно критиковать всех остальных, таких же «неудачников» как и я. + +Теперь давайте перестроим 16 подтипов Юнга через призму моего сознания, и назовем их соционика по Алвюнгу. Напомню, оригинал предлагает раскладывать людей на плоскости, где слева — мышление, справа — эмоции, сверху — интуиция, снизу — ощущение. Человек не может быть одновременно снизу и сверху, но может быть сверху и слева, например. Будем перестраивать, чтобы лучше охарактеризовать меня самого. + +Что я думаю по этому поводу? Во-первых, это красиво. А красота однозначно спасет мир. Во-вторых, в чем успех выбора определенной геометрии? Можно было выбрать треугольник, или шар, и натягивать людей на них. Почему плоскость (квадрат)? Думаю, в балансе между сложностью и банальностью спекулятивных историй о принадлежности. + +Тип 1. + + + + + + + + + + + + + + + + + + + + + + + + + + diff --git a/Essays/Речь на посвят фопф.md b/Essays/Речь на посвят фопф.md new file mode 100644 index 0000000..54e292a --- /dev/null +++ b/Essays/Речь на посвят фопф.md @@ -0,0 +1,13 @@ +Привет! + +Спасибо за приглашение, для меня это большая честь. + +Меня зовут Алвис. 15 лет назад я поступил на ФОПФ. Меня взяли, потому что я был членом национальный сборной Латвии по физике, так что выбор факультета был очевиден. Когда я первокурсником впервые взошел на эту сцену, нас готовили к посвящению, и попросили выстроиться полукругом. Мой однокурсник тогда поправил ведущих, что правильней сказать - полуэллипсом. Тогда я понял, что сделал правильный выбор. + +Сейчас я уже шестой год работаю техническим лидом в Хуавей. Выбор факультета здесь повлиял кардинально. Мой нанимающий был ФОПФ. Мой коллега в команде оказался парнем, которому я, будучи на третьем курсе главным куратором, во время чаепития на посвяте советовал не заедать антибиотики бутером из замороженной водки. Он, конечно, не прислушался. + +Моя команда меняла состав, но во все времена так получалось, что фопфы оказывались рядом. Я думаю так получается, потому что существует вселенское взаимное притяжение. Потому что после пройденной дороги, у выпускников одинаково выравнивается мировоззрение, на каком-то фундаментальном, душевном уровне. Работа с такими людьми заставляет любить свое дело. + +Но я конечно не хочу сказать, что одно лишь поступление на факультет - гарант успеха. Сильнейшие технические навыки - вот за что фопфов берут на работу прежде всего. Я хотел бы дать какие-то конкретные наставления, типа не забывайте добавлять константу интегрирования, но среди моих университетских друзей истории успеха настолько разные, что сложно сказать, что именно послужило ключевым навыком. Кто-то стал профессором биоинформатики в Гренобле, кто-то профессором физики в Лос Аламосе. Кто-то директором AI направления в крупной компании. Кто-то нашел себя в бизнесе и консалтинге. + +Поэтому я хотел бы вам пожелать найти свой путь, такой о котором вы бы никогда не смогли пожалеть, и который бы раскрыл ваш потенциал на максимум. Учитесь, будьте сильными, не расслабляйтесь и никогда не отчаивайтесь. Успехов вам на вашем пути! \ No newline at end of file diff --git a/Hello Note.md b/Hello Note.md new file mode 100644 index 0000000..33e30e0 --- /dev/null +++ b/Hello Note.md @@ -0,0 +1,10 @@ +This is a hello note. The first note in the new Personal Knowledge Base. + + +- The point of Daily is to revise and prioritize tasks, not to *complete* tasks. +- Split complex tasks and check deadlines. + +hello + +hello2 + diff --git a/Social/Social Circles.md b/Social/Social Circles.md new file mode 100644 index 0000000..f5ad837 --- /dev/null +++ b/Social/Social Circles.md @@ -0,0 +1,19 @@ +Dunbar’s number — Robin Dunbar (British anthropologist), based on neocortex-to-group-size correlation across primates. The famous figure is ~150 (stable social relationships), but Dunbar also proposed nested layers, each roughly ~3× the previous: + +• 5 — intimates / support clique (closest loved ones) +• 15 — sympathy group (close friends, people whose death would devastate you) +• 50 — friends (regular social contact) +• 150 — meaningful relationships / “casual friends” (the canonical Dunbar’s number) +• 500 — acquaintances +• 1500 — faces/names you can recognize + +The 150 figure comes from extrapolating the primate neocortex-ratio vs. group-size regression to humans. It’s widely cited but also contested — there have been replication critiques (Lindenfors et al. 2021 argued the confidence interval is enormous, anywhere from ~2 to ~520), so treat it as a useful heuristic rather than a hard constant. + +# Friends + + +Circle 3 - calls around half year. + +## Robert + +Last called 18.04.2026. Still Amazon, no changes. Talked about AI bubble. Invited to Russia. Wants to meet in Latvia when possible. Told him I’m married, he is not. Next call — around September. \ No newline at end of file