How to use this moduleCara menggunakan modul ini

You are an adult learner with your own goals, so you choose the path. Every section opens with why it matters, gives you a real problem first, and ends with a self-check that tells you immediately whether you are ready to move on. Tick "section done" at the end of each section to track progress on this device.

Anda pelajar dewasa dengan matlamat sendiri, jadi anda memilih laluan. Setiap seksyen dimulakan dengan mengapa ia penting, memberi masalah sebenar dahulu, dan diakhiri dengan semakan kendiri yang memberitahu dengan segera sama ada anda bersedia untuk meneruskan. Tandakan "seksyen selesai" di akhir setiap seksyen untuk menjejak kemajuan pada peranti ini.

Four-hour planPelan empat jam

§1 · Empathise & DefineEmpati & Takrif · 25 min · CLO1

The problem comes firstMasalah didahulukan

1.1 Empathy map (5 min)Peta empati (5 min)

For each stakeholder, write one line for says, thinks, feels, does. Then convert each pain into a constraint. Expected result: at least two hard constraints and three soft constraints.

Bagi setiap pihak berkepentingan, tulis satu baris untuk kata, fikir, rasa, buat. Kemudian tukarkan setiap kesakitan kepada kekangan. Hasil dijangka: sekurang-kurangnya dua kekangan keras dan tiga kekangan lembut.

1.2 Define: three ingredients of local searchTakrif: tiga ramuan carian setempat

1.3 Why hill climbing is not enoughMengapa pendakian bukit tidak mencukupi

Hill climbing accepts a neighbour only if it is at least as good. It stops at a local optimum: a solution with no better neighbour, although a better solution exists elsewhere (the global optimum). The landscape image helps: the height is solution quality, and the neighbourhood decides which points are "next to" each other. Change the neighbourhood and you change the landscape.

Pendakian bukit menerima jiran hanya jika ia sekurang-kurangnya sama baik. Ia berhenti pada optimum setempat: penyelesaian tanpa jiran yang lebih baik, walaupun penyelesaian lebih baik wujud di tempat lain (optimum sejagat). Imej landskap membantu: ketinggian ialah kualiti penyelesaian, dan kejiranan menentukan titik mana yang "bersebelahan". Tukar kejiranan dan anda menukar landskap.

Three classic escape strategies exist. This module covers the first one; L7 covers the second.

Tiga strategi klasik untuk melepasi wujud. Modul ini merangkumi yang pertama; L7 merangkumi yang kedua.

§2 · Ideate AJana idea A · 60 min · CLO1, CLO2

Simulated AnnealingPenyepuhlindapan Simulasi

Heat, then cool slowly. Accept a worse move with a probability that shrinks as the search cools.Panaskan, kemudian sejukkan perlahan. Terima langkah lebih buruk dengan kebarangkalian yang mengecil apabila carian menyejuk.

2.1 Where the idea comes fromAsal usul idea

Metallurgists anneal metal by heating it until atoms move freely and then cooling it slowly, so the atoms settle into a low-energy crystal. Cooling too fast (quenching) freezes defects in place. In 1953 Metropolis and colleagues proposed a rule to simulate such systems on a computer Metropolis et al. 1953. Thirty years later, Kirkpatrick, Gelatt and Vecchi Kirkpatrick et al. 1983, and independently Černý Černý 1985, turned the rule into a general optimiser and applied it to chip placement and the travelling salesman problem.

Ahli metalurgi menyepuhlindap logam dengan memanaskannya sehingga atom bergerak bebas, kemudian menyejukkannya perlahan-lahan supaya atom mendap ke hablur bertenaga rendah. Penyejukan terlalu cepat (lindap kejut) memerangkap kecacatan. Pada 1953, Metropolis dan rakan-rakan mencadangkan peraturan untuk menyimulasikan sistem sebegini pada komputer Metropolis et al. 1953. Tiga puluh tahun kemudian, Kirkpatrick, Gelatt dan Vecchi Kirkpatrick et al. 1983, dan secara berasingan Černý Černý 1985, menjadikan peraturan itu pengoptimum umum dan menggunakannya untuk penempatan cip dan masalah jurujual kembara.

Milestones: 1953 → 2026Tonggak: 1953 → 2026

2.2 The Metropolis acceptance rulePeraturan penerimaan Metropolis

For minimisation, let Δ = f(s′) − f(s). If Δ ≤ 0 the neighbour is no worse, so accept it. If Δ > 0, accept it with probability:

Bagi peminimuman, biar Δ = f(s′) − f(s). Jika Δ ≤ 0 jiran tidak lebih buruk, maka terima. Jika Δ > 0, terima dengan kebarangkalian:

Read the formula as two dials. A larger worsening Δ lowers P. A higher temperature T raises P. For maximisation use Δ = f(s) − f(s′) so that Δ is still positive for a worse move.

Baca formula sebagai dua tombol. Kemerosotan Δ yang lebih besar merendahkan P. Suhu T yang lebih tinggi menaikkan P. Bagi pemaksimuman gunakan Δ = f(s) − f(s′) supaya Δ masih positif bagi langkah lebih buruk.

2.3 Intuition in four cellsIntuisi dalam empat sel

Early in the run (hot) SA explores, accepting most moves. Late in the run (cold) it exploits and behaves like hill climbing. The cooling schedule moves it smoothly from one to the other.

Awal larian (panas) SA meneroka dan menerima kebanyakan langkah. Akhir larian (sejuk) ia mengeksploit dan berkelakuan seperti pendakian bukit. Jadual penyejukan menggerakkannya dengan lancar dari satu ke yang lain.

2.4 The process flow, then line by lineAliran proses, kemudian baris demi baris

Now the same logic as pseudocode. Hover or tab through the explanations to highlight the lines.

Kini logik yang sama sebagai pseudokod. Halakan tetikus atau tab melalui penerangan untuk menyerlahkan baris.

s ← initial solution;  best ← s
T ← T0
while T > Tmin and budget remains:
    repeat L times:
        s' ← random neighbour of s
        Δ ← f(s') − f(s)
        if Δ ≤ 0 or random() < exp(−Δ / T):
            s ← s'
            if f(s) < f(best): best ← s
    T ← α · T
return best

2.5 Cooling schedulesJadual penyejukan

2.6 Choosing the parametersMemilih parameter

2.7 Worked examples at three levelsContoh berpandu pada tiga tahap

2.8 Variations and the research frontierVariasi dan barisan hadapan penyelidikan

§2.9 · Lab 1Makmal 1 · 20 min · CLO2

Landscape LabMakmal Landskap

This lab maximises a one-dimensional landscape, so you can see every move. Two algorithms run side by side from the same start (x₀ on the far left) with the same number of steps. For GD the shaded area is the water; for SA watch the count of worse moves accepted.

Makmal ini memaksimumkan landskap satu dimensi supaya anda dapat melihat setiap langkah. Dua algoritma berjalan seiring dari titik mula yang sama (x₀ di sebelah kiri) dengan bilangan langkah yang sama. Bagi GD kawasan berlorek ialah air; bagi SA perhatikan bilangan langkah buruk yang diterima.

§3 · Ideate BJana idea B · 60 min · CLO1, CLO2

The Great DelugeAlgoritma Banjir Besar

Let the water rise and keep your feet dry. Accept any move that stays on the right side of a moving level.Biar air naik dan kaki tetap kering. Terima sebarang langkah yang kekal di sebelah betul paras yang bergerak.

3.1 The metaphorMetafora

Gunter Dueck described a hiker caught in endless rain Dueck 1993. The hiker wants the highest point but walks somewhat randomly, accepting any step that keeps their feet above the water. The water rises at a constant rate, so low ground disappears and the walker is pushed onto higher ground. Earlier, Dueck and Scheuer had shown that a simple deterministic threshold could match or beat SA Dueck & Scheuer 1990; the deluge makes that threshold move.

Gunter Dueck menggambarkan seorang pendaki yang terperangkap dalam hujan tanpa henti Dueck 1993. Pendaki mahukan titik tertinggi tetapi berjalan secara agak rawak, menerima sebarang langkah yang mengekalkan kakinya di atas air. Air naik pada kadar tetap, jadi tanah rendah hilang dan pejalan ditolak ke tanah lebih tinggi. Sebelumnya, Dueck dan Scheuer menunjukkan bahawa ambang berketentuan yang mudah boleh menyamai atau mengatasi SA Dueck & Scheuer 1990; banjir besar menjadikan ambang itu bergerak.

The "or not worse than current" part comes from the practical version by Burke, Bykov, Newall and Petrovic for exam timetabling Burke et al. 2004. Dueck's original accepts based on the level only.

Bahagian "atau tidak lebih buruk daripada semasa" datang daripada versi praktikal oleh Burke, Bykov, Newall dan Petrovic untuk penjadualan peperiksaan Burke et al. 2004. Versi asal Dueck menerima berdasarkan paras sahaja.

Milestones: 1990 → 2026Tonggak: 1990 → 2026

3.2 The process flow, then line by lineAliran proses, kemudian baris demi baris

s ← initial solution;  best ← s
B ← f(s)                          # level starts at the initial cost
ΔB ← (f(s) − f*) / N              # N = budget, f* = target
for k = 1 .. N:
    s' ← random neighbour of s
    if f(s') ≤ f(s) or f(s') ≤ B:
        s ← s'
        if f(s) < f(best): best ← s
    B ← B − ΔB
return best

3.3 Setting the decay rateMenetapkan kadar susutan

Two questions replace temperature tuning: How long can you run? (N) and What result would you accept? (f*). Burke and colleagues called this a time-predefined search: the user trades time for quality explicitly Burke et al. 2004.

Dua soalan menggantikan penalaan suhu: Berapa lama boleh dijalankan? (N) dan Keputusan apa yang boleh diterima? (f*). Burke dan rakan-rakan menamakannya carian masa pratakrif: pengguna menukar masa dengan kualiti secara jelas Burke et al. 2004.

3.4 Worked examples at three levelsContoh berpandu pada tiga tahap

3.5 The threshold familyKeluarga ambang

Try TA, RRT and LAHC in the Landscape Lab: choose them as Algorithm A or B.

Cuba TA, RRT dan LAHC dalam Makmal Landskap: pilih sebagai Algoritma A atau B.

3.6 SA and GD side by sideSA dan GD bersebelahan

Neither dominates. The No Free Lunch theorems say that averaged over all problems, no search algorithm beats another Wolpert & Macready 1997. What matters is the fit to your problem, and that is decided by testing (§4).

Tiada yang mendominasi. Teorem Tiada Makan Tengah Hari Percuma menyatakan bahawa dipuratakan ke atas semua masalah, tiada algoritma carian yang mengatasi yang lain Wolpert & Macready 1997. Yang penting ialah kesesuaian dengan masalah anda, dan itu ditentukan melalui pengujian (§4).

3.7 Close to home: timetabling researchDekat dengan kita: penyelidikan penjadualan

Timetabling is where deluge-style methods earned their reputation. The 2007 International Timetabling Competition set shared benchmarks McCollum et al. 2010, and Müller's winning solver combined constructive search with hill climbing, great deluge and simulated annealing Müller 2009. At FTSM UKM, researchers have used GD and SA for course and examination timetabling, for example GD with a Kempe-chain neighbourhood Shaker & Abdullah 2009, dual-sequence SA Abdullah et al. 2010, and a hybrid that integrates GD with other mechanisms for exam timetabling Turabieh & Abdullah 2011.

Penjadualan ialah bidang di mana kaedah gaya banjir membina reputasinya. Pertandingan Penjadualan Antarabangsa 2007 menetapkan penanda aras bersama McCollum et al. 2010, dan pemenang oleh Müller menggabungkan carian konstruktif dengan pendakian bukit, banjir besar dan penyepuhlindapan simulasi Müller 2009. Di FTSM UKM, penyelidik telah menggunakan GD dan SA untuk penjadualan kursus dan peperiksaan, contohnya GD dengan kejiranan rantai Kempe Shaker & Abdullah 2009, SA jujukan dwi Abdullah et al. 2010, dan hibrid yang mengintegrasikan GD dengan mekanisme lain untuk penjadualan peperiksaan Turabieh & Abdullah 2011.

§3.8 · Side by sideBersebelahan · 15 min · CLO1

Two flows, one skeletonDua aliran, satu rangka

SA and GD share most of their process. They differ in exactly three boxes: how they start, how they judge a worse move, and what they update after each move.SA dan GD berkongsi kebanyakan proses. Ia berbeza tepat pada tiga kotak: cara bermula, cara menilai langkah buruk, dan apa yang dikemas kini selepas setiap langkah.

3.8.1 Dual trace: same offers, two gatekeepersSurih dwi: tawaran sama, dua penjaga pintu

Both algorithms start at cost 120 and receive the same eight candidate costs. Step through one iteration at a time: the gold path shows the route each one takes through its flowchart. Gold columns in the strip mark iterations where they make different decisions. (In a real run the neighbours would differ once the two paths diverge; here we feed both the same costs so the gates can be compared directly.)

Kedua-dua algoritma bermula pada kos 120 dan menerima lapan kos calon yang sama. Langkah satu lelaran pada satu masa: laluan emas menunjukkan laluan setiap algoritma melalui carta alirnya. Lajur emas dalam jalur menandakan lelaran di mana keputusan berbeza. (Dalam larian sebenar jiran akan berbeza sebaik laluan berpisah; di sini kedua-duanya diberi kos yang sama supaya pintu dapat dibandingkan secara langsung.)

§3.9 · Lab 2Makmal 2 · 25 min · CLO2, CLO3

Route Lab: Klang ValleyMakmal Laluan: Lembah Klang

A mobile health-screening van leaves UKM Bangi (gold dot) and must visit 20 towns once and return: a travelling salesman problem. Distances are straight-line kilometres from approximate town coordinates. HC, SA and GD start from the same random tour, use the same 2-opt neighbourhood, and get the same number of evaluations.

Sebuah van saringan kesihatan bergerak bertolak dari UKM Bangi (titik emas) dan mesti melawat 20 bandar sekali dan kembali: masalah jurujual kembara. Jarak ialah kilometer garis lurus daripada koordinat anggaran bandar. HC, SA dan GD bermula dari lawatan rawak yang sama, menggunakan kejiranan 2-opt yang sama, dan mendapat bilangan penilaian yang sama.

§4 · Prototype & TestPrototaip & Uji · 25 min · CLO2, CLO3

Exam Timetable LabMakmal Jadual Peperiksaan

Back to the PBL problem, scaled down so you can see it: 12 papers, 8 slots (4 days × morning and afternoon) and 7 student groups. The data are invented for teaching. Cost = Σ over groups of (group size × penalty per student) + fairness weight × 100 × (penalty of the worst-off group). Penalty per student = 1000 × clashes + w_b × back-to-back pairs (two papers on the same day).

Kembali kepada masalah PBL, dikecilkan supaya anda dapat melihatnya: 12 kertas, 8 slot (4 hari × pagi dan petang) dan 7 kumpulan pelajar. Data direka untuk pengajaran. Kos = Σ bagi kumpulan (saiz kumpulan × penalti setiap pelajar) + pemberat keadilan × 100 × (penalti kumpulan paling terjejas). Penalti setiap pelajar = 1000 × pertindihan + w_b × pasangan berturutan (dua kertas pada hari yang sama).

Test honestly: a protocolUji dengan jujur: satu protokol

1. Same budget. Compare on number of evaluations (or the same wall-clock time on the same machine).
2. Bajet sama. Bandingkan pada bilangan penilaian (atau masa jam dinding yang sama pada mesin yang sama).
3. Same starts, many seeds. At least 5 seeds in class, 30 for a paper.
4. Permulaan sama, banyak benih. Sekurang-kurangnya 5 benih di kelas, 30 untuk kertas penyelidikan.
5. Report the distribution. Mean ± standard deviation, best, worst; a box plot if you can.
6. Laporkan taburan. Purata ± sisihan piawai, terbaik, terburuk; plot kotak jika boleh.
7. Test significance. For research, use a non-parametric test such as Wilcoxon signed-rank on paired runs.
8. Uji kesignifikanan. Untuk penyelidikan, gunakan ujian bukan parametrik seperti Wilcoxon pangkat bertanda pada larian berpasangan.
9. Tune fairly. Give every method the same tuning effort; a tuned SA against a default GD is not a fair comparison Johnson et al. 1989.
10. Tala secara adil. Beri setiap kaedah usaha penalaan yang sama; SA yang ditala lawan GD lalai bukan perbandingan adil Johnson et al. 1989.

§4.1 · Lab 4 · Emulation ArenaMakmal 4 · Arena Emulasi · 20 min · CLO1, CLO3, CLO4

Three problems, two gatekeepersTiga masalah, dua penjaga pintu

Run hill climbing, SA and GD on the same three problems, with the same budget and the same starting solutions, over many seeds. Watch how fast each one converges, how often it accepts worse moves, and where it fails. The diagnosis panel names each weakness and offers a fix you can try with one click.Jalankan pendakian bukit, SA dan GD pada tiga masalah yang sama, dengan bajet dan penyelesaian permulaan yang sama, atas banyak benih. Perhatikan kelajuan penumpuan, kekerapan menerima langkah buruk, dan di mana ia gagal. Panel diagnosis menamakan setiap kelemahan dan menawarkan pembaikan yang boleh dicuba dengan satu klik.

Strengths, weaknesses and potentialKekuatan, kelemahan dan potensi

§4.2 · Low-code toolkitKit alat kod rendah · 20 min · CLO3

Build it without programmingBina tanpa pengaturcaraan

Companion toolsAlat pendamping

Copy-ready prompts at three levelsProm sedia salin pada tiga tahap

Each prompt has a role, a task, constraints, an output format and a self-check. The explanation under each prompt tells you what every line does, so you can adapt it. Prompts are in English because most assistants follow technical instructions most reliably in English; the last line asks for the answer in formal Bahasa Malaysia when you need it.

Setiap prom mempunyai peranan, tugasan, kekangan, format output dan semakan kendiri. Penerangan di bawah setiap prom memberitahu fungsi setiap baris supaya anda boleh mengubah suainya. Prom ditulis dalam bahasa Inggeris kerana kebanyakan pembantu mengikut arahan teknikal paling tepat dalam bahasa Inggeris; baris terakhir meminta jawapan dalam Bahasa Malaysia formal apabila diperlukan.

You are a patient tutor for postgraduate students in Artificial Intelligence.
Explain Simulated Annealing and the Great Deluge algorithm.
Use one everyday analogy for each, then state each acceptance rule as a formula.
Keep it under 300 words and avoid code.
End with three multiple-choice questions (with answers) that test the difference between the two.
Write the answer in formal Bahasa Malaysia (no mixing with English), keeping symbols such as T, Δ and B.

SA ≈ metal cooled slowly; P(terima) = e^(−Δ/T) …
GD ≈ hujan dan paras air; terima jika f(s′) ≤ B, B ← B − ΔB …
S1. Algoritma manakah tidak menggunakan nombor rawak dalam keputusan penerimaan? … Jawapan: GD

Check my simulated annealing trace for a minimisation problem.
Settings: T0 = 10, alpha = 0.9, cool after every move, accept worse if r < exp(-delta/T).
Rows (k, f(s), f(s'), r): (1,50,46,-), (2,46,49,0.52), (3,49,57,0.61), (4,49,44,-), (5,44,47,0.70)
For each row, show T, delta, P = exp(-delta/T) to 4 decimals, and accept or reject.
Show the arithmetic so I can verify it. Do not skip rows.
At the end, list any row where my expected decision (accept, accept, reject, accept, reject) disagrees with yours.

k=1 T=10.0000 Δ=−4 → accept (better)
k=2 T=9.0000  Δ=3  P=0.7165 r=0.52 → accept
k=3 T=8.1000  Δ=8  P=0.3724 r=0.61 → reject
k=4 T=7.2900  Δ=−5 → accept (better)
k=5 T=6.5610  Δ=3  P=0.6330 r=0.70 → reject
No disagreements.

Act as a reviewer for an operations-research journal.
I am designing a great deluge search for a university exam timetable.
Anonymised summary only: 40 papers, 12 slots, about 1,200 students in 7 cohorts (C1–C7), no names or IDs.
Cost = 1000 x clashes + 10 x back-to-back + 3 x same-day, summed over students.
Budget N = 200,000 evaluations; initial cost 54,000; target f* = 3,000.
1) Is the decay rate sensible? Show the value of delta-B.
2) Identify two risks in this objective for small cohorts, and propose one fairness term.
3) Propose a fair experiment to compare against simulated annealing.
Mark anything you are unsure about as UNSURE, and do not invent citations.

1) ΔB = 0.255 per evaluation; sensible if 3,000 is reachable …
2) Risks: headcount weighting hides small cohorts; one 1000-point clash dominates …
   Fairness term: + w · max over cohorts of penalty per student …
3) Same N, same initial timetables, 30 seeds each, Wilcoxon signed-rank …

Sources: Dueck (1993) and Burke et al. (2004).
Build a comparison table: acceptance rule, parameters, how the level is initialised and updated, test problems, and main finding.
Quote the sentence from each paper that defines the acceptance rule, with its location.
List one limitation each paper admits, and one it does not mention.

Simulated Annealing in a spreadsheetPenyepuhlindapan Simulasi dalam hamparan

Maximise f(x) = x·sin(x) on [0, 10] starting from x = 2, which is a local peak (f ≈ 1.82). The global peak is at x ≈ 7.98 (f ≈ 7.92). Put the parameters in B1:B3, the headers in row 6, the formulas below in row 7 (start) and row 8, then fill row 8 down to row 506. Works in Google Sheets and Excel.

Maksimumkan f(x) = x·sin(x) pada [0, 10] bermula dari x = 2, iaitu puncak setempat (f ≈ 1.82). Puncak sejagat pada x ≈ 7.98 (f ≈ 7.92). Letak parameter dalam B1:B3, pengepala dalam baris 6, formula di bawah dalam baris 7 (permulaan) dan baris 8, kemudian isi baris 8 ke bawah hingga baris 506. Berfungsi dalam Google Sheets dan Excel.

B1: 10            (T0)
B2: 0.99          (alpha)
B3: 1             (step size)
Row 6 headers: k | T | x | x_new | f(x) | f(x_new) | delta | P | r | accept | best
A7: 0             B7: =$B$1          C7: 2
D7: =MIN(10,MAX(0,C7+NORM.INV(RAND(),0,$B$3)))
E7: =C7*SIN(C7)   F7: =D7*SIN(D7)    G7: =E7-F7
H7: =IF(G7<=0,1,EXP(-G7/B7))         I7: =RAND()
J7: =IF(I7
    

      
1–3Parameters in one place, so learners can change them without touching formulas.Parameter di satu tempat supaya pelajar boleh mengubahnya tanpa menyentuh formula.
      
5Start state: iteration 0, temperature T₀, x = 2 (the trap).Keadaan mula: lelaran 0, suhu T₀, x = 2 (perangkap).
      
6Neighbour: Gaussian step, clipped to [0, 10].Jiran: langkah Gauss, dipotong ke [0, 10].
      
7Objective values and Δ = f(x) − f(x_new): positive means worse (we maximise).Nilai objektif dan Δ = f(x) − f(x_new): positif bermaksud lebih buruk (kita memaksimumkan).
      
8Metropolis probability and the random draw.Kebarangkalian Metropolis dan nombor rawak.
      
9Accept flag and best-so-far.Bendera terima dan terbaik setakat ini.
      
10Next row: count up, cool geometrically, move only if accepted.Baris seterusnya: kira naik, sejukkan secara geometri, bergerak hanya jika diterima.
      
11Fill down 500 rows. Press F9 (Excel) or edit any cell (Sheets) to rerun.Isi ke bawah 500 baris. Tekan F9 (Excel) atau sunting mana-mana sel (Sheets) untuk jalankan semula.
    
    
Expected outputOutput dijangka
K506 (best f) ≈ 7.917 in roughly 3 of every 4 recalculations
(our simulation of this exact sheet: 78% of 3,000 reruns).
Change H7 to =IF(G7<=0,1,0) (hill climbing): best stays 1.820 almost every time (<1%).
Raise B2 to 0.995 and fill to row 1006: success rises to about 95%.
  

Optional: Python (for those who want it)Pilihan: Python (bagi yang berminat)

import math, random

def f(x):                       # objective to MAXIMISE on [0, 10]
    return x * math.sin(x)

def neighbour(x, step=1.0):     # small random move, kept inside [0, 10]
    return min(10, max(0, x + random.gauss(0, step)))

def hill_climb(x, iters=1000):
    for _ in range(iters):
        y = neighbour(x)
        if f(y) >= f(x):
            x = y
    return x

def simulated_annealing(x, T0=10.0, alpha=0.995, iters=1000):
    T, best = T0, x
    for _ in range(iters):
        y = neighbour(x)
        delta = f(x) - f(y)                     # > 0 means y is worse
        if delta <= 0 or random.random() < math.exp(-delta / T):
            x = y
        if f(x) > f(best):
            best = x
        T *= alpha                              # geometric cooling
    return best

def great_deluge(x, level=-6.0, target=7.9, iters=1000):
    rain = (target - level) / iters             # how fast the water rises
    best = x
    for _ in range(iters):
        y = neighbour(x)
        if f(y) >= f(x) or f(y) >= level:
            x = y
        if f(x) > f(best):
            best = x
        level += rain
    return best

random.seed(42)
for name, algo in [("HC", hill_climb), ("SA", simulated_annealing), ("GD", great_deluge)]:
    wins = 0
    for run in range(100):
        xb = algo(2.0)
        wins += f(xb) > 7.5
    print(f"{name}: reached the global peak in {wins}/100 runs; last run x={xb:.3f}, f={f(xb):.3f}")

HC: reached the global peak in 1/100 runs; last run x=2.028, f=1.820
SA: reached the global peak in 93/100 runs; last run x=2.029, f=1.820
GD: reached the global peak in 91/100 runs; last run x=7.315, f=6.277

import numpy as np
from scipy.optimize import dual_annealing

def cost(v):                      # 2-D Rastrigin: many local minima, global minimum 0 at (0, 0)
    return 20 + sum(x**2 - 10*np.cos(2*np.pi*x) for x in v)

result = dual_annealing(cost, bounds=[(-5.12, 5.12)] * 2, seed=42)
print("x* =", np.round(result.x, 4))
print("f(x*) =", round(result.fun, 6))
print("evaluations:", result.nfev)

x* = [-0. -0.]
f(x*) = 0.0
evaluations: 4133

§4.3 · Safety, ethics & data protectionKeselamatan, etika & perlindungan data · 15 min · CLO4

Optimise responsiblyOptimumkan secara bertanggungjawab

An optimiser does exactly what the objective says, at scale. Two kinds of risk follow: risk to people's data when we use tools, and risk to people's outcomes when we deploy the result.

Pengoptimum melakukan tepat apa yang dinyatakan objektif, pada skala besar. Dua jenis risiko menyusul: risiko kepada data individu apabila kita menggunakan alat, dan risiko kepada hasil yang diterima individu apabila kita menggunakan keputusan.

Scenarios: what would you do?Senario: apa yang anda akan lakukan?

§5 · Applications & frontierAplikasi & barisan hadapan · 15 min · CLO1, CLO4

Choose by applicationPilih mengikut aplikasi

5.1 A decision treePokok keputusan

Rules of thumb, not laws. The test step (§4.1) decides.Panduan umum, bukan undang-undang. Langkah ujian (§4.1) yang menentukan.

5.2 Your problem: an interactive chooserMasalah anda: pemilih interaktif

Tick what is true for your problem (your research, a client's scheduling task, or the PBL timetable). The map shows which known applications sit closest to your answer.

Tandakan yang benar bagi masalah anda (penyelidikan anda, tugasan penjadualan klien, atau jadual PBL). Peta menunjukkan aplikasi yang paling hampir dengan jawapan anda.

5.3 Where SA and GD work todayDi mana SA dan GD bekerja hari ini

§6 · Reflect & assessRenung & nilai · 25 min · CLO1–CLO4

Final ChallengeCabaran Akhir

Twelve questions mixing recall, calculation and judgement. Feedback appears immediately. Aim for 9/12 or better; if you score lower, the explanation tells you which section to revisit.

Dua belas soalan yang menggabungkan ingatan, pengiraan dan pertimbangan. Maklum balas muncul serta-merta. Sasarkan 9/12 atau lebih; jika skor lebih rendah, penerangan memberitahu seksyen yang perlu diulang kaji.

Exit ticket · 3-2-1Tiket keluar · 3-2-1

• 3 things you learned today.
• 3 perkara yang anda pelajari hari ini.
• 2 places in your work or research where SA or GD could help.
• 2 tempat dalam kerja atau penyelidikan anda di mana SA atau GD boleh membantu.
• 1 question that is still open for you.
• 1 soalan yang masih terbuka bagi anda.

PBL deliverable (group, due next week)Hasil PBL (berkumpulan, perlu dihantar minggu depan)

§7 · Further reading & resourcesBacaan lanjut & sumber

Keep goingTeruskan

Glossary · BM–ENGlosari · BM–EN

§8

ReferencesRujukan

DOIs were checked against Crossref/publisher records on 30 Oct 2026. Policy items link to official or law-firm summaries; confirm details with the official text before relying on them.DOI disemak terhadap rekod Crossref/penerbit pada 30 Okt 2026. Item dasar dipautkan kepada ringkasan rasmi atau firma guaman; sahkan butiran dengan teks rasmi sebelum bergantung padanya.

1. Abdullah, S., Shaker, K., McCollum, B. & McMullan, P. (2010). Dual sequence simulated annealing with round-robin approach for university course timetabling. In EvoCOP 2010, LNCS, Springer, 1–10. https://doi.org/10.1007/978-3-642-12139-5_1
2. Ahir, S., Bhat, P. & Doboli, A. (2026). HYPERHEURIST: A simulated annealing-based control framework for LLM-driven code generation in optimized hardware design. arXiv:2604.15642 (preprint). https://arxiv.org/abs/2604.15642
3. Aramon, M., Rosenberg, G., Valiante, E., Miyazawa, T., Tamura, H. & Katzgraber, H. G. (2019). Physics-inspired optimization for quadratic unconstrained problems using a digital annealer. Frontiers in Physics, 7, 48. https://doi.org/10.3389/fphy.2019.00048
4. Baykasoğlu, A., Subulan, K. & Hamzadayı, A. (2026). Great deluge-based metaheuristic incorporating integer nonlinear programming for modeling and solving dynamic capability-based machine layout problem. Computers & Operations Research, 185, 107302. https://doi.org/10.1016/j.cor.2025.107302
5. Borders, W. A., Pervaiz, A. Z., Fukami, S., Camsari, K. Y., Ohno, H. & Datta, S. (2019). Integer factorization using stochastic magnetic tunnel junctions. Nature, 573, 390–393. https://doi.org/10.1038/s41586-019-1557-9
6. Burke, E. K. & Bykov, Y. (2017). The late acceptance hill-climbing heuristic. European Journal of Operational Research, 258(1), 70–78. https://doi.org/10.1016/j.ejor.2016.07.012
7. Burke, E. K. & Kendall, G. (Eds.) (2014). Search Methodologies: Introductory Tutorials in Optimization and Decision Support Techniques (2nd ed.). Springer. https://doi.org/10.1007/978-1-4614-6940-7
8. Burke, E. K., Bykov, Y., Newall, J. & Petrovic, S. (2004). A time-predefined local search approach to exam timetabling problems. IIE Transactions, 36(6), 509–528. https://doi.org/10.1080/07408170490438410
9. Chen, S. (2022, March). This Month in Physics History: March 1953 — the Metropolis Monte Carlo algorithm. APS News, American Physical Society. https://www.aps.org/apsnews/2022/03/rosenbluth-metropolis-monte-carlo-algorithm
10. Cheng, C.-K., Kahng, A. B., Kundu, S., Wang, Y. & Wang, Z. (2023). Assessment of reinforcement learning for macro placement. Proc. ISPD '23, ACM, 158–166. https://doi.org/10.1145/3569052.3578926
11. Correia, A. H. C., Worrall, D. E. & Bondesan, R. (2023). Neural simulated annealing. Proc. AISTATS 2023, PMLR 206, 4946–4962. https://proceedings.mlr.press/v206/correia23a.html
12. Delahaye, D., Chaimatanan, S. & Mongeau, M. (2019). Simulated annealing: From basics to applications. In M. Gendreau & J.-Y. Potvin (Eds.), Handbook of Metaheuristics (3rd ed.), Springer, 1–35. https://doi.org/10.1007/978-3-319-91086-4_1
13. Dueck, G. & Scheuer, T. (1990). Threshold accepting: A general purpose optimization algorithm appearing superior to simulated annealing. Journal of Computational Physics, 90(1), 161–175. https://doi.org/10.1016/0021-9991(90)90201-B
14. Dueck, G. (1993). New optimization heuristics: The great deluge algorithm and the record-to-record travel. Journal of Computational Physics, 104(1), 86–92. https://doi.org/10.1006/jcph.1993.1010
15. Dueck, G. (2004). Das Sintflutprinzip: Ein Mathematik-Roman. Springer. https://openlibrary.org/works/OL9074734W
16. Earl, D. J. & Deem, M. W. (2005). Parallel tempering: Theory, applications, and new perspectives. Physical Chemistry Chemical Physics, 7(23), 3910–3916. https://doi.org/10.1039/b509983h
17. Franzin, A. & Stützle, T. (2019). Revisiting simulated annealing: A component-based analysis. Computers & Operations Research. https://doi.org/10.1016/j.cor.2018.12.015
18. Geman, S. & Geman, D. (1984). Stochastic relaxation, Gibbs distributions, and the Bayesian restoration of images. IEEE Transactions on Pattern Analysis and Machine Intelligence, PAMI-6(6), 721–741. https://doi.org/10.1109/TPAMI.1984.4767596
19. Goto, H., Tatsumura, K. & Dixon, A. R. (2019). Combinatorial optimization by simulating adiabatic bifurcations in nonlinear Hamiltonian systems. Science Advances, 5(4), eaav2372. https://doi.org/10.1126/sciadv.aav2372
20. Hajek, B. (1988). Cooling schedules for optimal annealing. Mathematics of Operations Research, 13(2), 311–329. https://doi.org/10.1287/moor.13.2.311
21. Hastings, W. K. (1970). Monte Carlo sampling methods using Markov chains and their applications. Biometrika, 57(1), 97–109. https://doi.org/10.1093/biomet/57.1.97
22. HPCwire (2011, 26 May). D-Wave sells first quantum computer. https://www.hpcwire.com/2011/05/26/d-wave_sells_first_quantum_computer/
23. Hutama, R. R. & Muklason, A. (2024). Adaptive learning modified great deluge hyper-heuristics. International Journal of Computing, 23(2), 287–293. https://doi.org/10.47839/ijc.23.2.3549
24. Ingber, L. (1989). Very fast simulated re-annealing. Mathematical and Computer Modelling, 12(8), 967–973. https://doi.org/10.1016/0895-7177(89)90202-1
25. Johnson, D. S., Aragon, C. R., McGeoch, L. A. & Schevon, C. (1989). Optimization by simulated annealing: An experimental evaluation; Part I, graph partitioning. Operations Research, 37(6), 865–892. https://doi.org/10.1287/opre.37.6.865
26. Kadowaki, T. & Nishimori, H. (1998). Quantum annealing in the transverse Ising model. Physical Review E, 58(5), 5355–5363. https://doi.org/10.1103/PhysRevE.58.5355
27. King, A. D., Nocera, A., Rams, M. M., Dziarmaga, J., Wiersema, R., et al. (2025). Beyond-classical computation in quantum simulation. Science, 388(6743), 199–204. https://doi.org/10.1126/science.ado6285
28. King, A. D., Raymond, J., Lanting, T., et al. (2023). Quantum critical dynamics in a 5,000-qubit programmable spin glass. Nature, 617, 61–66. https://doi.org/10.1038/s41586-023-05867-2
29. Kirkpatrick, S., Gelatt, C. D. & Vecchi, M. P. (1983). Optimization by simulated annealing. Science, 220(4598), 671–680. https://doi.org/10.1126/science.220.4598.671
30. Landa-Silva, D. & Obit, J. H. (2008). Great deluge with non-linear decay rate for solving course timetabling problems. Proc. 4th International IEEE Conference on Intelligent Systems (IS'08), Varna, 8-11–8-18. https://doi.org/10.1109/IS.2008.4670447
31. Liu, F., Tong, X., Yuan, M., et al. (2024). Evolution of heuristics: Towards efficient automatic algorithm design using large language model. Proc. ICML 2024, PMLR 235. https://arxiv.org/abs/2401.02051
32. Lundy, M. & Mees, A. (1986). Convergence of an annealing algorithm. Mathematical Programming, 34(1), 111–124. https://doi.org/10.1007/BF01582166
33. McCollum, B., Schaerf, A., Paechter, B., McMullan, P., Lewis, R., Parkes, A. J., Di Gaspero, L., Qu, R. & Burke, E. K. (2010). Setting the research agenda in automated timetabling: The second international timetabling competition. INFORMS Journal on Computing, 22(1), 120–130. https://doi.org/10.1287/ijoc.1090.0320
34. McMullan, P. (2007). An extended implementation of the great deluge algorithm for course timetabling. In Computational Science – ICCS 2007, LNCS, Springer, 538–545. https://doi.org/10.1007/978-3-540-72584-8_71
35. Metropolis, N., Rosenbluth, A. W., Rosenbluth, M. N., Teller, A. H. & Teller, E. (1953). Equation of state calculations by fast computing machines. Journal of Chemical Physics, 21(6), 1087–1092. https://doi.org/10.1063/1.1699114
36. Ministry of Science, Technology and Innovation (2024). National Guidelines on AI Governance and Ethics (AIGE). Government of Malaysia. https://www.malaysia.gov.my/en/my-initiative/whole-government-digital-services/tadbir-urus/national-guidelines-on-artificial-intelligence-governance-and-ethics-aige
37. Munoz-Bauza, H. & Lidar, D. (2025). Scaling advantage in approximate optimization with quantum annealing. Physical Review Letters, 134(16), 160601. https://doi.org/10.1103/physrevlett.134.160601
38. Müller, T. (2009). ITC2007 solver description: A hybrid approach. Annals of Operations Research, 172(1), 429–446. https://doi.org/10.1007/s10479-009-0644-y
39. National AI Office (2024). Press release: Launch of the National AI Office, 12 December 2024. https://ai.gov.my/media/news-details/20241212_Press_Release_Launch_of_NAIO.pdf
40. Nikhar, S., Kannan, S., Aadit, N. A., Chowdhury, S. & Camsari, K. Y. (2024). All-to-all reconfigurability with sparse and higher-order Ising machines. Nature Communications, 15, 8977. https://doi.org/10.1038/s41467-024-53270-w
41. Novikov, A., Vũ, N., Eisenberger, M., et al. (2025). AlphaEvolve: A coding agent for scientific and algorithmic discovery. arXiv:2506.13131. https://arxiv.org/abs/2506.13131
42. Personal Data Protection (Amendment) Act 2024, Malaysia: phased commencement 1 Jan, 1 Apr and 1 Jun 2025 (summary by HHQ). https://hhq.com.my/posts/full-implementation-of-the-personal-data-protection-amendment-act-2024-final-roundup/
43. Qiu, N. & Liang, D. (2025). Reinforcement learning based simulated annealing. Proc. AAMAS 2025. https://aamas.csc.liv.ac.uk/Proceedings/aamas2025/pdfs/p1718.pdf
44. Regulation (EU) 2024/1689 (Artificial Intelligence Act). Official Journal of the European Union. https://eur-lex.europa.eu/eli/reg/2024/1689/oj
45. Regulation (EU) 2026/1744 (Digital Omnibus on AI). Official Journal of the European Union. https://eur-lex.europa.eu/eli/reg/2026/1744/oj
46. Romera-Paredes, B., Barekatain, M., Novikov, A., et al. (2024). Mathematical discoveries from program search with large language models. Nature, 625, 468–475. https://doi.org/10.1038/s41586-023-06924-6
47. Russell, S. & Norvig, P. (2022). Artificial Intelligence: A Modern Approach (4th ed., Global Edition). Pearson. https://aima.cs.berkeley.edu/
48. Sechen, C. & Sangiovanni-Vincentelli, A. (1985). The TimberWolf placement and routing package. IEEE Journal of Solid-State Circuits, 20(2), 510–522. https://doi.org/10.1109/JSSC.1985.1052337
49. Shaker, K. & Abdullah, S. (2009). Incorporating great deluge approach with Kempe chain neighbourhood structure for curriculum-based course timetabling problems. 2nd Conference on Data Mining and Optimization (DMO), Kajang, 149–153. https://doi.org/10.1109/DMO.2009.5341894
50. Swendsen, R. H. & Wang, J.-S. (1986). Replica Monte Carlo simulation of spin-glasses. Physical Review Letters, 57(21), 2607–2609. https://doi.org/10.1103/PhysRevLett.57.2607
51. Talbi, E.-G. (2009). Metaheuristics: From Design to Implementation. John Wiley & Sons. https://doi.org/10.1002/9780470496916
52. The Star (2026, 11 September). AI governance bill expected by early 2027, says Gobind. https://www.thestar.com.my/news/nation/2026/09/11/ai-governance-bill-expected-by-early-2027-says-gobind
53. Tsallis, C. & Stariolo, D. A. (1996). Generalized simulated annealing. Physica A, 233(1–2), 395–406. https://doi.org/10.1016/S0378-4371(96)00271-3
54. Turabieh, H. & Abdullah, S. (2011). An integrated hybrid approach to the examination timetabling problem. Omega, 39(6), 598–607. https://doi.org/10.1016/j.omega.2010.12.005
55. UNESCO (2021). Recommendation on the Ethics of Artificial Intelligence. https://www.unesco.org/en/artificial-intelligence/recommendation-ethics
56. van Stein, N. & Bäck, T. (2025). LLaMEA: A large language model evolutionary algorithm for automatically generating metaheuristics. IEEE Transactions on Evolutionary Computation, 29(2), 331–345. https://doi.org/10.1109/TEVC.2024.3497793
57. Wolpert, D. H. & Macready, W. G. (1997). No free lunch theorems for optimization. IEEE Transactions on Evolutionary Computation, 1(1), 67–82. https://doi.org/10.1109/4235.585893
58. Xiang, Y., Sun, D. Y., Fan, W. & Gong, X. G. (1997). Generalized simulated annealing algorithm and its application to the Thomson model. Physics Letters A, 233(3), 216–220. https://doi.org/10.1016/S0375-9601(97)00474-X
59. Yang, C., Wang, X., Lu, Y., Liu, H., Le, Q. V., Zhou, D. & Chen, X. (2024). Large language models as optimizers. ICLR 2024. https://arxiv.org/abs/2309.03409
60. Ye, H., Wang, J., Cao, Z., et al. (2024). ReEvo: Large language models as hyper-heuristics with reflective evolution. NeurIPS 2024. https://arxiv.org/abs/2402.01145
61. Černý, V. (1985). Thermodynamical approach to the traveling salesman problem: An efficient simulation algorithm. Journal of Optimization Theory and Applications, 45(1), 41–51. https://doi.org/10.1007/BF00940812