Upgrade to Pro
— share decks privately, control downloads, hide ads and more …
Speaker Deck
Features
Speaker Deck
PRO
Sign in
Sign up for free
Search
Search
EliasFano
Search
Shunsuke Kanda
November 30, 2019
Research
320
1
Share
Embed
Copy iframe code
Copy JS code
Copy link
Start on current slide
EliasFano
10th StringBeginnersでの発表資料
Shunsuke Kanda
November 30, 2019
More Decks by Shunsuke Kanda
See All by Shunsuke Kanda
Leveraging LLMs for Unsupervised Dense Retriever Ranking (SIGIR 2024)
kampersanda
3
490
Lucene/Elasticsearch の Character Filter でユニコード正規化するとトークンのオフセットがズレるバグへの Workaround - Search Engineering Tech Talk 2024 Spring
kampersanda
0
1.6k
Binary and Scalar Embedding Quantization for Significantly Faster & Cheaper Retrieval
kampersanda
3
560
トライとダブル配列の基礎
kampersanda
2
1.8k
Binary search with modern processors
kampersanda
34
15k
AIP Open Seminar #6
kampersanda
0
310
ICDM2020
kampersanda
0
300
SIGSPATIAL20
kampersanda
0
280
Fast Succinct Trie
kampersanda
2
790
Other Decks in Research
See All in Research
National high-resolution cropland classification of Japan with agricultural census information and multi-temporal multi-modality datasets
satai
3
470
2026年度 生成AI を活用した論文執筆ガイド/ワークショップ / 2026 Academic Year Guide to Writing Papers Using Generative AI - Workshop
ks91
PRO
0
220
Visual SLAM未来予測 / Future Prediction in Visual SLAM
koide3
1
940
IA for theory
gpeyre
1
380
進学校の生徒にはア行の苗字が多いのか
ozekinote
0
550
最先端NLP 2026 論文紹介: Wait, Wait, Wait... Why Do Reasoning Models Loop? / SNLP Paper Review: Wait, Wait, Wait... Why Do Reasoning Models Loop?
tkng
0
140
データサイエンティストの就労意識~2015 → 2026 一般(個人)会員アンケートより
datascientistsociety
PRO
0
700
Harness Engineering and Al Agent
kzinmr
3
1.9k
[BlackHatAsia2026] Hidden Telemetry: Uncovering TraceLogging ETW Providers You're Not Using (Yet)
asuna_jp
1
680
Karkada さんの論文 × 2 の紹介: (1) Closed-Form Training Dynamics Reveal Learned Features and Linear Structure in Word2Vec-like Models, (2) Symmetry in language statistics shapes the geometry of model representations
eumesy
PRO
1
340
The story of RefactoringMiner. Slow research, long-term impact
tsantalis
0
120
最先端NLP勉強会2026 論文紹介:Reasoning with Sampling: Your Base Model is Smarter Than You Think (ICLR 2026 paper)
kogoro
3
400
Featured
See All Featured
Amusing Abliteration
ianozsvald
1
270
30 Presentation Tips
portentint
PRO
1
380
Pawsitive SEO: Lessons from My Dog (and Many Mistakes) on Thriving as a Consultant in the Age of AI
davidcarrasco
0
220
How to build a perfect <img>
jonoalderson
1
5.9k
Evolution of real-time – Irina Nazarova, EuRuKo, 2024
irinanazarova
9
1.5k
The Pragmatic Product Professional
lauravandoore
37
7.4k
Documentation Writing (for coders)
carmenintech
77
5.5k
The innovator’s Mindset - Leading Through an Era of Exponential Change - McGill University 2025
jdejongh
PRO
1
300
Effective software design: The role of men in debugging patriarchy in IT @ Voxxed Days AMS
baasie
0
490
Sharpening the Axe: The Primacy of Toolmaking
bcantrill
46
3k
Principles of Awesome APIs and How to Build Them.
keavy
128
18k
Bioeconomy Workshop: Dr. Julius Ecuru, Opportunities for a Bioeconomy in West Africa
akademiya2063
PRO
1
300
Transcript
EliasFano 10th StringBeginners → Kanda →
K and A → K & A → K ampersand A →
⟫ × "$& n '% S[0,n) × i.e.,
S[i-1] ≤ S[i] for each 0 < i < n ⟫ ! × Access / Predecessor / Successor ( × #( 2 0 1 2 3 4 5 6 7 S 2 7 18 28 42 43 44 59 Access(4) = 42 Predecessor(10) = 7 Successor(43) = 44
EliasFano 3 S 2 7 18 28 42 43 44
59 000 000 010 011 101 101 101 111 n = 8 u log $ log % & 010 111 010 100 010 011 100 011 2 1 3 1 1 001 100 110 0 0 0 log H = 110 0 10 10 0 1110 0 10 2 0 1 1 0 0 1 3 ( ) L = 010 111 010 100 010 011 100 011 ()
2" + " log ' ( 4 L
= 010 111 010 100 010 011 100 011 H = 110 0 10 10 0 1110 0 10 0 1 2 3 4 5 6 7 S 2 7 18 28 42 43 44 59 EliasFano log ) " " log ' ( " 2" 2 *+, ( ≈ "
2" + " log ' ( "6 ⟫
[0,u) 2 n (,/ (i.e. 4) -) #. % $*0+3 6 5 S 2 4 5 8 9 0010110011 2 45 89 n 5'& 2 u ' ( 1 5!5 3 log ' (
2" + " log ' ( 4 ⟫
[0,u) 1 n *- +& !, # "'.(2 4 6 S 2 4 4 5 8 9 0010011010001010 2 5 ')( ( / 33 (0)) 44 8 9 n 3%$1 u+n Less than half a bit per element away (Quasi-succinct) 2 log ')( ( ≈ " log ')( (
⟫ Access(i) = S[i] ⟫ Predecessor(x) = max{S[i] :
S[i] ≤ x} ⟫ Successor(x) = min{S[i] : S[i] > x} 7 O(1) O(log $ % ) 0 1 2 3 4 5 6 7 S 2 7 18 28 42 43 44 59 Access(4) = 42 Predecessor(10) = 7 Successor(43) = 44
Access&O(1) $ ⟫ &Access(4) = 42 = 101 0102 8
L = 010 111 010 100 010 011 100 011 H = 110 0 10 10 0 1110 0 10 ① !"# $ % i ! Select1 (4) – 4 = 9 – 4 = 5 = 1012 ② !"# % Select1 (i) – i H Select " o(n) Selectb (H, i)&H i ! b # %
SuccessorO(log $ % ) ⟫ Successor(43) = 44 =
101 1002 9 L = 010 111 010 100 010 011 100 011 H = 110 0 10 10 0 1110 0 10 Select0 (4) = 8 Select0 (5) = 12 log $ % = 3 3 × (Select0 (4) – 4) = 12 3 × (Select0 (5) – 5) = 21 101 0112 5 ② O(&'( ) * ) ① &'( * Select
/6 EliasFano7) ⟫ # (;+' AA ×
10%?*& × =98. " × TRIE 2! × 3<5,4!@: -> 10 10 7 5 0 1 4 8 6 9 3 a t e t a a t e c c 2 $A