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
Beyond Accuracy: Behavioral Testing of NLP Mode...
Search
Scatter Lab Inc.
July 31, 2020
Research
2.4k
0
Share
Embed
Copy iframe code
Copy JS code
Copy link
Start on current slide
Beyond Accuracy: Behavioral Testing of NLP Models with CheckList
Scatter Lab Inc.
July 31, 2020
More Decks by Scatter Lab Inc.
See All by Scatter Lab Inc.
zeta introduction
scatterlab
0
2k
SimCLR: A Simple Framework for Contrastive Learning of Visual Representations
scatterlab
0
4.5k
Adversarial Filters of Dataset Biases
scatterlab
0
2.3k
Sparse, Dense, and Attentional Representations for Text Retrieval
scatterlab
0
2.4k
Weight Poisoning Attacks on Pre-trained Models
scatterlab
0
2.2k
Approximate Nearest Neighbor Negative Contrastive Learning for Dense Text Retrieval
scatterlab
0
2.6k
Open-Retrieval Conversational Question Answering
scatterlab
0
2.3k
What Can Neural Networks Reason About?
scatterlab
0
2.3k
Exploring the Limits of Transfer Learning with Unified Text-to-Text Transformer
scatterlab
0
2.3k
Other Decks in Research
See All in Research
Spatial Active Noise Control Based onSound Field Interpolation Incorporating Physical Constraints
skoyamalab
0
160
National high-resolution cropland classification of Japan with agricultural census information and multi-temporal multi-modality datasets
satai
3
450
Fukui Shibiten 39 - AI Art
butchi
0
180
GLIM とMegaParticles:正規分布近似の限界とタイトカップリング&パーティクルフィルタの進展 / GLIM and MegaParticles : Progress of the distribution representation in SLAM
koide3
0
720
計算情報学研究室(数理情報学第7研究室)2026
tomohirokoana
0
770
Google Cloud Next 2026 DM Recap Agentic Data Cloudを添えて / Google Cloud Next 2026 DM Recap
nnaka2992
0
120
人間中心の意思決定支援AI
yukinobaba
PRO
7
3.8k
Vector Map as Language: Toward Unified Remote Sensing Vector Mapping
satai
3
190
Dual Quadric表現を用いた動的物体追跡とRGB-D・IMU制約の密結合によるオドメトリ推定
nanoshimarobot
0
510
[最先端NLP勉強会2026] Checklists Are Better Than Reward Models For Aligning Language Models
nzw0301
1
140
敵対生成プロンプト同時探索による内省型プロンプト最適化
kinoue_smarthr
0
370
The story of RefactoringMiner. Slow research, long-term impact
tsantalis
0
110
Featured
See All Featured
Imperfection Machines: The Place of Print at Facebook
scottboms
270
14k
Impact Scores and Hybrid Strategies: The future of link building
tamaranovitovic
0
420
Navigating Team Friction
lara
192
16k
Save Time (by Creating Custom Rails Generators)
garrettdimon
PRO
32
4.4k
A designer walks into a library…
pauljervisheath
211
24k
Effective software design: The role of men in debugging patriarchy in IT @ Voxxed Days AMS
baasie
0
480
"I'm Feeling Lucky" - Building Great Search Experiences for Today's Users (#IAC19)
danielanewman
230
23k
Noah Learner - AI + Me: how we built a GSC Bulk Export data pipeline
techseoconnect
PRO
0
390
Building Adaptive Systems
keathley
44
3.2k
Odyssey Design
rkendrick25
PRO
2
770
We Analyzed 250 Million AI Search Results: Here's What I Found
joshbly
1
1.8k
実際に使うSQLの書き方 徹底解説 / pgcon21j-tutorial
soudai
PRO
201
75k
Transcript
#FZPOE"DDVSBDZ#FIBWJPSBM5FTUJOHPG/-1 .PEFMTXJUI$)&$,-*45 ߅ള .-3FTFBSDI4DJFOUJTU 1JOHQPOH
0WFSWJFX 0WFSWJFX "$- #FTU1BQFS .JDSPTPGU3FTFBSDI 6OJWPG8BTIJOHUPO 6OJWPG$BMJGPSOJB
*SWJOF $PEFIUUQTHJUIVCDPNNBSDPUDSDIFDLMJTU ೠਃড নೠ/-1ݽ؛ਸ4PGUXBSF5FTUJOHҙীࢲಣо೧ࠁ
ݾର ݾର *OUSPEVDUJPO $)&$,-*45 5FTUJOH405"XJUI$)&$,-*45 $PNNFSDJBM4ZTUFNBOE$)&$,-*45
$PODMVTJPO
*OUSPEVDUJPO *OUSPEVDUJPO
*OUSPEVDUJPO (FOFSBMJ[BUJPOPG/-1NPEFMT ↟ ੌ߈ച (FOFSBMJ[BUJPO ח٩۞ݽ؛୶ҳ೧ঠೞחоݾೞաੑפ ↟ ݽ؛णೠؘఠࡺ݅ইפۄ ীחೠߣبࠁঋؘఠ VOTFFOEBUB
ীࢲبੜ೧ঠפ ↟ ݽ؛ੌ߈ചػࢿמਸࡅܰѱঈೞӝਤ೧ ؘఠࣇਸള۲ಣоࣇਵ۽ҳ࠙ ↟ ୭Ӕীח)VNBOQFSGPNBODFܳڪযֈחݽ؛ٜبࣻઓ
*OUSPEVDUJPO (FOFSBMJ[BUJPOPG/-1NPEFMT ↟ ੌ߈ച (FOFSBMJ[BUJPO ח٩۞ݽ؛୶ҳ೧ঠೞחоݾೞաੑפ ↟ ݽ؛णೠؘఠࡺ݅ইפۄ ীחೠߣبࠁঋؘఠ VOTFFOEBUB
ীࢲبੜ೧ঠפ ↟ ݽ؛ੌ߈ചػࢿמਸࡅܰѱঈೞӝਤ೧ ؘఠࣇਸള۲ಣоࣇਵ۽ҳ࠙ ↟ ୭Ӕীח)VNBOQFSGPNBODFܳڪযֈחݽ؛ٜبࣻઓ ↟ ೞ݅ಣоؘఠীࢲੜೞחѪਵ۽࠙ೡө ↟ ٩۞ݽ؛ਵ۽যܳ೧ೞӝࠁח 5BTLTQFDJGJDೞҊ%BUBTQFDJGJDೠಁఢਸ৻ਛӝٸޙীࢿמ֫חبԲળઁӝؾפ ↟ ٮۄࢲഅۢMFBEFSCPBSE֫Ҕਸೱೠোҳߑೱ ஞݽ؛מ۱ਸPWFSFTUJNBUFೡࣻبणפ ↟ tղо݅ٚݽ؛SFBMXPSMEীࢲઁ۽زೞחഛੋೞ۰ݶ যڌѱ೧ঠೡө u
(PPHMF4FOUJNFOU"OBMZTJT"1* *OUSPEVDUJPO
(PPHMF4FOUJNFOU"OBMZTJT"1* *OUSPEVDUJPO ↟ য়טݡधաࢁঋওݶযب/FVUSBMա৬ঠೞ݅ ৈ/FHBUJWFੑפ
(PPHMF4FOUJNFOU"OBMZTJT"1* *OUSPEVDUJPO
(PPHMF4FOUJNFOU"OBMZTJT"1* *OUSPEVDUJPO ↟ /FVUSBMޙ/FHBUJPO/FVUSBMৈঠೞ݅ ъೠ/FHBUJWFоա৳णפ
1SFWJPVT8PSL *OUSPEVDUJPO ↟ ഛب৻নೠߑߨਵ۽ݽ؛מ۱ਸಣоೞחোҳоઓפ ↟ بਵ۽য়ఋٜܳ݅যࢲݽ؛3PCVTUOFTTܳಣоೞӝ ↟ ޙਸQBSBQISBTJOH೧ࢲ߸ೞחࠁӝ ↟ ౠݺࢎܳജ೧ࢲݽ؛'BJSOFTTܳಣоೞӝ
tղҊೱೠҴঠu tղҊೱࣗ۲ঠu tղҊೱইܻঠu ↟ ӝઓোҳೠ҅ ↟ ࠗ࠙োҳח2" TFOUJNFOUBOBMZTJT /-*١ѐ߹ੋUBTL࢚ܳਵ۽ೣ ↟ ݽ؛ীೠನҚੋಣоо٘ۄੋغӝয۰
$)&$,-*45 $)&$,-*45
ઁউೞחߑߨݽ؛ب4PGUXBSFܳ5FTUೞחѪۢಣо೧ࠁ $)&$,-*45 ↟ 4PGUXBSF&OHJOFFSJOH࠙ঠt#FIBWJPS5FTUJOHu BLB#MBDLCPY5FTUJOH җਬࢎೠߑߨਸઁউפ
ઁউೞחߑߨݽ؛ب4PGUXBSFܳ5FTUೞחѪۢಣо೧ࠁ ↟ 4PGUXBSF&OHJOFFSJOH࠙ঠt#FIBWJPS5FTUJOHu BLB#MBDLCPY5FTUJOH җਬࢎೠߑߨਸઁউפ ↟ ղࠗҳઑחഃҊ۰ೞঋҊ ٯੑ۱۱݅оҊݽ؛מ۱ $BQBCJMJUZ ਸಣоפ
8IBUUP5FTU ↟ 5BTLীҙ҅হы୶যঠೞחৈ۞-JOHVJTUJD$BQBCJMJUZ FH /FHBUJPO /&3 ܳݽ؛ઁ۽ы୶Ҋחಣоפ $)&$,-*45
ઁউೞחߑߨݽ؛ب4PGUXBSFܳ5FTUೞחѪۢಣо೧ࠁ ↟ 4PGUXBSF&OHJOFFSJOH࠙ঠt#FIBWJPS5FTUJOHu BLB#MBDLCPY5FTUJOH җਬࢎೠߑߨਸઁউפ ↟ ղࠗҳઑחഃҊ۰ೞঋҊ ٯੑ۱۱݅оҊݽ؛מ۱ $BQBCJMJUZ ਸಣоפ
8IBUUP5FTU ↟ 5BTLীҙ҅হы୶যঠೞחৈ۞-JOHVJTUJD$BQBCJMJUZ FH /FHBUJPO /&3 ܳݽ؛ઁ۽ы୶Ҋחಣоפ ↟ ೞա$BQBCJMJUZоGBJMVSF೮ਸٸ ܳTQFDJGJDCFIBWJPS۽ա־ӝਤ೧ѐ5FTU5ZQFਸبੑפ )PXUP5FTU ↟ .JOJNVN'VODUJPOBMJUZ5FTU .'5 ݽ؛۱࢚чҗزੌೠ 6OJU5FTU ↟ *OWBSJBODF5FTU */7 ݽ؛ੑ۱߄Շযب۱җزੌೠ 1FSUVSCBUJPO5FTU ↟ %JSFDUJPOBM&YQFDUBUJPO5FTU %&5 ݽ؛ੑ۱ਸ߄Բݶ۱࢚чҗزੌೠ 1FSUVSCBUJPO5FTU $)&$,-*45
ઁউೞחߑߨݽ؛ب4PGUXBSFܳ5FTUೞחѪۢಣо೧ࠁ ↟ 4PGUXBSF&OHJOFFSJOH࠙ঠt#FIBWJPS5FTUJOHu BLB#MBDLCPY5FTUJOH җਬࢎೠߑߨਸઁউפ ↟ ղࠗҳઑחഃҊ۰ೞঋҊ ٯੑ۱۱݅оҊݽ؛מ۱ $BQBCJMJUZ ਸಣоפ
8IBUUP5FTU ↟ 5BTLীҙ҅হы୶যঠೞחৈ۞-JOHVJTUJD$BQBCJMJUZ FH /FHBUJPO /&3 ܳݽ؛ઁ۽ы୶Ҋחಣоפ ↟ ೞա$BQBCJMJUZоGBJMVSF೮ਸٸ ܳTQFDJGJDCFIBWJPS۽ա־ӝਤ೧ѐ5FTU5ZQFਸبੑפ )PXUP5FTU ↟ .JOJNVN'VODUJPOBMJUZ5FTU .'5 ݽ؛۱࢚чҗزੌೠ 6OJU5FTU ↟ *OWBSJBODF5FTU */7 ݽ؛ੑ۱߄Շযب۱җزੌೠ 1FSUVSCBUJPO5FTU ↟ %JSFDUJPOBM&YQFDUBUJPO5FTU %&5 ݽ؛ੑ۱ਸ߄Բݶ۱࢚чҗزੌೠ 1FSUVSCBUJPO5FTU ↟ ݆పझ٘ܳࡅܰѱ݅٘חѪਸذӝਤ೧UFNQMBUF MFYJDPO QFSUVSCBUJPO WJTVBMJ[BUJPO .BTLFE-.١5PPMਸઁҕפ $)&$,-*45
&YBNQMF ↟ ಣоೞח$BQBCJMJUZ/&("5*0/ ↟ 5FTU5ZQF.'5 ۱࢚җزੌೠ $)&$,-*45
&YBNQMF ↟ ಣоೞח$BQBCJMJUZ/&3 ↟ 5FTU5ZQF*/7 ੑ۱׳ۄઉب۱زੌೠ $)&$,-*45
&YBNQMF ↟ ಣоೞח$BQBCJMJUZ7PDBCVMBSZ ↟ 5FTU5ZQF%*3 ੑ۱׳ۄٸ۱࢚җزੌೠ $)&$,-*45
&YBNQMF $)&$,-*45 ↟ ৬эపझоաݶ ਤۢݽ؛ыযמ۱ਸಣоೡࣻणפ
6TBHF&YBNQMFమ݁࠼ਸࣚਵ۽ਕࢲపझாझٜܳ݅ӝ $)&$,-*45
6TBHF&YBNQMFమ݁࠼ਸ-FYJDPOਵ۽ਕࢲపझாझٜܳ݅ӝ $)&$,-*45
6TBHF&YBNQMFమ݁࠼ਸ#&35.BTLFE-.ਵ۽ਕࢲపझாझٜܳ݅ӝ $)&$,-*45
6TBHF&YBNQMFమ݁࠼ਸ#&35.BTLFE-.ਵ۽ਕࢲపझாझٜܳ݅ӝ $)&$,-*45
5FTUJOH405"XJUI$)&$,-*45 5FTUJOH405"XJUI$)&$,-*45
405"ݽ؛ٜਸಣо೧ࠁ 5FTUJOH405"XJUI$)&$,-*45 ↟ Ӕؘࢎपখ5BTL݅ࠁݶ ޙઁоցޖएਕࠁੌࣻبणפ ↟ ѐ5BTLী೧೯פ ↟ 4FOUJNFOU"OBMZTJT ↟
%VQMJDBUFE2VFTUJPO%FUFDUJPO ↟ 3FBEJOH$PNQSFIFOTJPO ↟ ৈ۞ݽ؛ٜਸ࢚ਵ۽פ ↟ $PNNFSDJBMNPEFMT(PPHMF .4 "NB[PO ↟ 405"NPEFMT#&35 3P#&35B
4FOUJNFOU"OBMZTJT 5FTUJOH405"XJUI$)&$,-*45 ↟ /FVUSBM8PSEܳ߄Ծীب4FOUJNFOUѾҗо߄Շणפ
4FOUJNFOU"OBMZTJT 5FTUJOH405"XJUI$)&$,-*45 ↟ ࢎѤࢲࣽਸ߈ೞޅೣਸঌࣻणפ
4FOUJNFOU"OBMZTJT 5FTUJOH405"XJUI$)&$,-*45 ↟ ױࣽೠࠗޙীبपಁೠѾҗо݆णפ
4FOUJNFOU"OBMZTJT 5FTUJOH405"XJUI$)&$,-*45 ↟ খޙਸߣࠂೞחޙࠢਵݶஏѾҗо߄Շযঠೞ݅ Ӓ۞ޅפ
%VQMJDBUFE2VFTUJPO%FUFDUJPO 5FTUJOH405"XJUI$)&$,-*45
%VQMJDBUFE2VFTUJPO%FUFDUJPO 5FTUJOH405"XJUI$)&$,-*45 ↟ Ѿҗஏীҙ҅হחݺࢎоജغݶѾҗо߄Շযߡ݀פ
3FBEJOH$PNQSFIFOTJPO 5FTUJOH405"XJUI$)&$,-*45 ↟ IBNTUFSWFIJDMFۄחҙ҅ܳঈೞޅפ
3FBEJOH$PNQSFIFOTJPO 5FTUJOH405"XJUI$)&$,-*45 ↟ য়ఋ١QFSUVSCBUJPOীхೞѱ߈פ
$PNNFSDJBM4ZTUFNBOE $)&$,-*45 $PNNFSDJBM4ZTUFNBOE$)&$,-*45
$PNNFSDJBM4ZTUFNBOE$)&$,-*45 ↟ খীࢲࠅࣻ٠ *5ӝসࢲ࠺झա405"ݽ؛ۄبষ݆ߡӒоߊѼغणפ ↟ Ӓۢ$)&$,-*45оߡӒܳחؘ݃աبؼࣻਸө
$PNNFSDJBM4ZTUFNBOE$)&$,-*45 6TFSTUVEZ#&35۽݅ٚ%21ܳदрزউಣо೧ࠁ ↟ ೠਃড $)&$,-*45оਵݶ؊݆పझாझٜܳ݅Ҋ ؊݆ߡӒܳওणפ $BQBCJMJUZ$BQBCJMJUZ 5FNQMBUF
$PODMVTJPO $PODMVTJPO
$PODMVTJPO ↟ /-1ݽ؛ਸ4PGUXBSFۢಣо೧ঠೠ ↟ Ӓۧݶޖਸ যڌѱಣо೧ঠೡө ↟ ޖਸ-JOHVJTUJD$BQBCJMJUZ ↟
যڌѱ6OJU5FTU .'5 1FSUVSCBUJPOT */7 %*3 ↟ ܳذӝਤೠ.BTLFE-. दпച -FYJDPO١ਸઁҕפ ↟ $PEFIUUQTHJUIVCDPNNBSDPUDSDIFDLMJTU $PODMVTJPO
хࢎפ✌ ୶оޙژחҾӘೠݶઁٚইېোۅ۽োۅࣁਃ ߅ള .-3FTFBSDI4DJFOUJTU 1JOHQPOH &NBJMDIBFIVO!TDBUUFSMBCDPLS