Kelly Marchisio (St. Denis) @cheeesio Twitter profile

Last Seen Profiles

@gpnvg18

@PChichidashi

@mats_lav

@fishcoooo

@Cleaverszn

@StalkingVi96051

@RCTtheatres

@SharkyMarc

@avosmanzabun

@CooperMays

@Sumiayu3

@itxLilyana

@RichShorney

@Solbliminal

@MariMilehamIHES

@PauBobii

@Levicknb

@EcHistSoc

@spider4554

@MSACL

@JayLovesPotato_

@tennis_insights

@drmngnow

@bindassrakhii

@StalkingVi96051

@zoronoamarimoo

@PasutriAsli2

@herywrb

@CTeshima

@stw_pdg

@FanboyEint

@merv8888

@JimDeshaies

@hoesloveszaddy

@Sarah96039017

Kelly Marchisio (St. Denis)

@cheeesio

2 months

Train steps: [34500/40000] (86.3%) GPU utilization: 100% Saving checkpoint: 7BB_marchisio_stdenis_pretrain_mixture_v0_34500.ckpt

59

135

3K

Kelly Marchisio (St. Denis)

@cheeesio

4 years

Freshly added to my .bash_profile: alias cdd="cd ../.." alias cddd="cd ../../.." alias cdddd="cd ../../../.." I'm not lazy, I'm... efficient... right??

18

16

240

Kelly Marchisio (St. Denis)

@cheeesio

3 months

How does quantization affect multilingual LLMs? 🌍 For wide adoption, multilingual LLMs must be highly-performant *and* lightweight. 📈 🪶 We analyze SOTA multilingual LLMs in 23 languages under various quantization techniques to find out! 📜

9

60

241

Kelly Marchisio (St. Denis)

@cheeesio

10 months

(mostly positive!) Reflections as a female AI researcher at #NeurIPS2023 part 1 of ?? A man from unnamed-but-very-well-known-AI-company approached me near the booth. He asked about my research and biggest challenges in multilingual NLP. A rapidfire back-and-forth ensued: 1/5 🧵

3

11

192

Kelly Marchisio (St. Denis)

@cheeesio

3 years

Just trained an MT model. The output for every test sentence is: " I & amp ; apos ; m sorry . I & amp ; apos ; m sorry . I & amp ; apos ; m sorry . I & amp ; apos ; m sorry . I & amp ; apos ; m sorry ." It's not your fault, little buddy! It's me, not you!

4

3

184

Kelly Marchisio (St. Denis)

@cheeesio

1 year

Life update! Tomorrow, I join @CohereAI as a Member of Technical Staff!

14

4

182

Kelly Marchisio (St. Denis)

@cheeesio

10 months

I was pleased to be treated as an equal, and for the opportunity to sharpen my intellectual battle sword⚔️🤺 (and proud that I was deffo 💯💯💯 correct 🤪💁‍♀️🏋️‍♀️🏄‍♀️🕵️‍♀️😜) 5/5

1

146

Kelly Marchisio (St. Denis)

@cheeesio

2 years

I'm on the job market! (industry/post-doc/faculty) I work on multilinguality and low-resource NLP, with a focus on computational efficiency. Please don’t hesitate to reach out with opportunities (DM/email)! Applying broadly, flexible location! 🏖️❄️🏔️🌴☔️

3

33

109

Kelly Marchisio (St. Denis)

@cheeesio

2 years

New year, new name! I'll still publish as "Kelly Marchisio", but socially, you can call me "Kelly St. Denis" :)

8

0

110

Kelly Marchisio (St. Denis)

@cheeesio

3 years

Engagement posts are sooo cliché -- so we trained a neural language model to write ours: Engaged Engaged Engaged for her big big big move on the big fella hasn't even when we were top-notch! from the happiest and her unravel and her *<expletive>* today🤣🤣🤣🤣 #princesscut

7

3

102

Kelly Marchisio (St. Denis)

@cheeesio

5 months

24

0

99

Kelly Marchisio (St. Denis)

@cheeesio

1 year

Done!!

JHU CLSP

@jhuclsp

1 year

Congratulations to Kelly Marchisio @cheeesio (advised by Philipp Koehn) on successfully defending her @JHUCompSci PhD thesis "Multilinguality from Embedding Spaces: Algorithmic, Geometric and Data Considerations." Kelly will join @CohereAI @HopkinsEngineer

0

2

28

20

2

97

Kelly Marchisio (St. Denis)

@cheeesio

2 months

@mayhewsw Expected date to run first inference is Sept 2 - we’re currently setting up our eval suite, so remains to be seen. I have high hopes for this one

2

0

85

Kelly Marchisio (St. Denis)

@cheeesio

2 years

Introducing ✨Mini-Model Adaptation✨ - a new parameter- and compute-efficient method for rapid adaptation of pretrained models to new languages! 🧵1/5

3

8

77

Kelly Marchisio (St. Denis)

@cheeesio

3 years

Me: "Hm, I don't know much about X. I should watch a video." First comment on said video: 😳

3

1

70

Kelly Marchisio (St. Denis)

@cheeesio

2 years

Thesis-writing starts TODAY. Join me on my thesis-writing journey for a PhD in Computer Science / AI / ML / NLP! 1/N

3

2

63

Kelly Marchisio (St. Denis)

@cheeesio

10 months

But they don’t with me, and I don’t with them. I have brilliant female computer scientist friends, we just don’t tend to engage with each other this way. I left thinking “WOAH that was aggressive! But he’d do the same if I were male.” 4/5

3

2

62

Kelly Marchisio (St. Denis)

@cheeesio

4 months

DONE done. 👩‍🎓

JHU CLSP

@jhuclsp

4 months

Congrats to CLSP’ers graduating this year! 🥳🥳 Photo credit: @esalesk

0

12

90

3

1

61

Kelly Marchisio (St. Denis)

@cheeesio

6 months

At @cohere , we prioritize multilinguality!

Aidan Gomez

@aidangomez

6 months

Multilinguality is something that is crucial for equitable utility of this technology. We want our models to work for as many people, organizations, and markets as possible. We perform strongly across 10 languages and we're eager to expand this further.

3

5

91

1

3

58

Kelly Marchisio (St. Denis)

@cheeesio

2 years

Honored to be named an Amazon Fellow!

Amazon Science

@AmazonScience

2 years

Amazon and @HopkinsEngineer announced the first PhD fellowships and faculty research awards recipients as part of the JHU + Amazon Initiative for Interactive AI. Learn why Alexa AI VP @natarajan_prem says these projects will help drive new advances in AI. #ArtificalIntelligence

2

7

38

3

4

57

Kelly Marchisio (St. Denis)

@cheeesio

3 months

We all get a little *confused* sometimes 🫢🫨😵‍💫 - joint work with @seb_ruder @weiyinko_ml Alex Bérard, Théo Dehaze, hot off the press! ♨️

Sebastian Ruder

@seb_ruder

3 months

Understanding and Mitigating Language Confusion 😵‍💫 User: ¿De qué trata nuestro artículo? LLM: We analyze one of LLMs’ most jarring errors: their failure to generate text in the user’s desired language. 📑 💻

5

43

182

6

7

55

Kelly Marchisio (St. Denis)

@cheeesio

6 months

Our prioritization of multilinguality extends even to our tokenizer. Better tokenization -> better representations -> better cost-efficiency for you! 💸

Aidan Gomez

@aidangomez

6 months

One subtlety worth mentioning is how significant the tokenizer is to the cost to use models in non-english languages. Our tokenizer is meaningfully better than others at the 9 non-English languages, achieving up to a 2x effective cost reduction to use.

5

13

124

1

6

56

Kelly Marchisio (St. Denis)

@cheeesio

2 years

GBO (qualifying exam) passed! 🎉-> now, 😴🍕🍦 (it's like gym/tan/laundry, but for newly-minted PhD candidates👩‍💻)

7

0

53

Kelly Marchisio (St. Denis)

@cheeesio

2 years

The ability to extract accurate translation dictionaries from monolingual embedding spaces depends critically on their geometric similarity--"degree of isomorphism." We address this root-cause of faulty X-lingual mapping with ✨IsoVec✨ #EMNLP2022 🧵1/N

3

7

49

Kelly Marchisio (St. Denis)

@cheeesio

10 months

He left, then returned for sources. I left, pleased for standing my ground, & that the “battle” had *happened*. Let me explain: Respectful intellectual argument is a valuable skill to be honed. My male colleagues do it to each other. They hang, they banter & spar, they hack. 3/5

1

48

Kelly Marchisio (St. Denis)

@cheeesio

3 years

Hilarious that this pops up now while I’m at EMNLP. Nine years ago - I’d coded my first “hello world” only about 6 weeks earlier - my my how things have changed! 💻 💕 🤓

2

1

44

Kelly Marchisio (St. Denis)

@cheeesio

4 years

Our new unsupervised MT work is up on Arxiv

cs.CL Papers

@arxiv_cs_cl

4 years

When Does Unsupervised Machine Translation Work?. (arXiv:2004.05516v1 []) #NLProc

0

11

21

0

8

41

Kelly Marchisio (St. Denis)

@cheeesio

6 months

We released our best multilingual LLM yet, with support for 10 languages and open weights for research! Check it out! 🌍🌏🌎

Cohere For AI

@CohereForAI

6 months

C4AI Command R+ is a state-of-the-art RAG-optimized model with advanced tool use to automate sophisticated tasks, including multi-hop tool use. ✨ Command R+ is optimized for general reasoning and excels at multilingual performance evaluated across 10 languages. 🌎

1

2

22

1

3

40

Kelly Marchisio (St. Denis)

@cheeesio

2 years

Headed to ✨NAACL 2022 ✨tomorrow! Looking forward to an exciting week of chats about multilinguality and low-resource MT. Come say “hi” if you see me!

3

2

40

Kelly Marchisio (St. Denis)

@cheeesio

5 months

please show me the training data 🙃

0

1

39

Kelly Marchisio (St. Denis)

@cheeesio

2 years

Might supervised and unsupervised MT be mutually-beneficial? In our #NAACL2022 work, we ask whether the training methods result in systematically different output beyond what is visible via quality metrics like adequacy or BLEU. 🧵1/4

3

5

37

Kelly Marchisio (St. Denis)

@cheeesio

10 months

Arrived in New Orleans for #NeurIPS2023 ! I’ll be at the Cohere booth tomorrow (Mon) 2:30-3:30pm, and 9-11:30am Tues-Thurs - come by if you want to chat about anything and everything multilingual NLP!

0

3

36

Kelly Marchisio (St. Denis)

@cheeesio

7 months

So thrilled to show you what we’ve been working on!!

Aidan Gomez

@aidangomez

7 months

⌘-R Introducing Command-R, a model focused on scalability, RAG, and Tool Use. We've also released the weights for research use, we hope they're useful to the community!

31

186

1K

0

36

Kelly Marchisio (St. Denis)

@cheeesio

1 year

Took a break from thesis-writing on Monday to visit @esalesk at JHU's Edible Book Festival, presenting her edible rendition of our advisor's book! Cake recipe generated with Bard! 🤓 @jhuclsp

6

3

34

Kelly Marchisio (St. Denis)

@cheeesio

9 months

New year, new manager! So excited to work with you, @seb_ruder ! Folks - come join us!

Sebastian Ruder

@seb_ruder

9 months

I'm excited to announce that I've joined @cohere to help make LLMs more multilingual! It’s crazy how the capabilities of NLP models have evolved over the last years. I’m thrilled to work with a team full of smart, dedicated and kind individuals to push the boundaries of LLMs.

60

24

848

1

0

34

Kelly Marchisio (St. Denis)

@cheeesio

1 year

To appear in Findings of ACL2023! @artetxem @PSH_Lewis @yihong_thu

Kelly Marchisio (St. Denis)

@cheeesio

2 years

Introducing ✨Mini-Model Adaptation✨ - a new parameter- and compute-efficient method for rapid adaptation of pretrained models to new languages! 🧵1/5

3

8

77

3

4

33

Kelly Marchisio (St. Denis)

@cheeesio

2 years

From Monday until early October, I'll be interning with @artetxem at Meta AI in London. If you'll be in 🇬🇧 in the next few months, let's meet up!

0

1

31

Kelly Marchisio (St. Denis)

@cheeesio

10 months

… Him: “I don’t believe that” Me: cites sources … Him: “With infinite computation, that’s not true” Me: “Sure, but we live in reality. Infinity isn’t real.” … etc. etc. etc. 2/5

1

31

Kelly Marchisio (St. Denis)

@cheeesio

3 months

Very fun to work with @johnamqdang on this project!

John Dang

@johnamqdang

3 months

Is RLHF effective for aligning multilingual LLMs? 🤔 Our work studies multilingual preference optimization to train a new SOTA multilingual LLM, advancing the frontier of alignment techniques to 23 languages covering half the world’s population 🌎! 🧵 📜

15

54

180

0

2

29

Kelly Marchisio (St. Denis)

@cheeesio

3 months

(1) Automatic metrics severely underestimate damage from quantization. ⚠️ While automatic evals estimate deterioration of a quantized model relative to FP16 across tasks at a modest −0.3% for French and −1.7% for Japanese. Humans report the drops as −16.6% and −16.0% 👎👎

1

5

27

Kelly Marchisio (St. Denis)

@cheeesio

1 year

I’m here in Toronto! #ACL2023NLP I’ll present Mini-Model Adaptation in a virtual poster session tomorrow 11:00–12:30 Toronto time, and again in-person at the RepL4NLP workshop on Thursday. Come say 👋!! @yihong_thu @PSH_Lewis @artetxem

Kelly Marchisio (St. Denis)

@cheeesio

1 year

To appear in Findings of ACL2023! @artetxem @PSH_Lewis @yihong_thu

3

4

33

0

6

26

Kelly Marchisio (St. Denis)

@cheeesio

2 years

Life-hack: I watched a 3-minute YouTube video about steaming milk and now I’ve been complimented at the office two days in a row and called a “pro.” Please spam me other ways I can fool others into thinking I’m competent in 3mins or less!!! (Voilà ☕️)

6

0

22

Kelly Marchisio (St. Denis)

@cheeesio

2 years

It’s my decade codeaversary! Right around this time 10 years ago, I coded my first line: a “hello world” in C. My life has never been the same 🥰

2

0

22

Kelly Marchisio (St. Denis)

@cheeesio

2 months

I'll be presenting our recent work "How Does Quantization Affect Multilingual LLMs?" at the Cohere4AI ML Efficiency Group on Friday at noon Eastern (GMT-4). Come join in on the fun! To join, please fill out the form:

SreeHarsh (C#)

@Sree_Harsha_N

2 months

At the ML efficiency group, excited to have @cheeesio to present work on 'How does quantization affect multilingual LLMs'. Quantization is ever present in the large model stack -- but it can have unintended impacts on quality. Join in to find out :)

2

1

20

0

3

22

Kelly Marchisio (St. Denis)

@cheeesio

2 years

Low-resourcedness + domain/script shift + noise dramatically ⬇️⬇️ geometric similarity of word embedding spaces. #EMNLP2022 We improve BLI on non-isomorphic spaces using a new optimal transport-based graph-matching algorithm. 9am Sunday in Abu Dhabi! 1/4🧵

2

3

22

Kelly Marchisio (St. Denis)

@cheeesio

2 years

Public Service Announcement!! Watch this: Python2: round(1.5) -> 2.0 round(2.5) -> 3.0 Cool. Python3: round(1.5) -> 2. round(2.5) -> … 2. What?!?! (And the type change, too!) 🧵 1/3

2

5

22

Kelly Marchisio (St. Denis)

@cheeesio

2 years

I'll be presenting two papers starting 30mins from now at #EMNLP2022 ! ✨IsoVec✨ (below) as a poster, and 📈BLI... using Graph Matching via Optimal Transport 📉 (tweeting yesterday) live in Hall B! Join me! @jhuclsp @n_verma1 @AliSaadEldin @kevinduh Carey Priebe, Philipp Koehn

Kelly Marchisio (St. Denis)

@cheeesio

2 years

The ability to extract accurate translation dictionaries from monolingual embedding spaces depends critically on their geometric similarity--"degree of isomorphism." We address this root-cause of faulty X-lingual mapping with ✨IsoVec✨ #EMNLP2022 🧵1/N

3

7

49

0

1

21

Kelly Marchisio (St. Denis)

@cheeesio

1 year

Day 22-30ish: The full draft is complete! A few hours per week turned into all-day-every-day for a week or two, between adding intro/abstract/future work/conclusion, and making requested edits from my committee. I defend *tomorrow* at 2pm Eastern at JHU!

0

20

Kelly Marchisio (St. Denis)

@cheeesio

3 years

Can unsupervised MT be useful for high-resource language pairs? Joint work with @markuseful @GrangierDavid out on arxiv:

On Systematic Style Differences between Unsupervised and...

Modern unsupervised machine translation (MT) systems reach reasonable translation quality under clean and controlled data conditions. As the performance gap between supervised and unsupervised MT...

arxiv.org

1

3

19

Kelly Marchisio (St. Denis)

@cheeesio

2 months

It's our final push on improving multilingual MMLU! If you speak any of the languages below, please consider contributing!

Cohere For AI

@CohereForAI

2 months

With 1⃣ week left in our MMLU Translation sprint, we are 22% through the task. ⌛️ Korean, Arabic, Vietnamese, Amharic, German, Indonesian, Chinese, Sinhala, Nepali, and Swedish are all closing in on the goal! 🥅 🌎 Speak these languages? Join us:

0

8

21

0

5

18

Kelly Marchisio (St. Denis)

@cheeesio

3 months

(2) Languages are disparately affected by quantization: non-Latin script languages are impacted worst 🥺 We knew they were poorly represented in training data & tokenization, causing ⏬ performance and ⏫ cost/latency. Now we know they’re treated unfairly in quantization, too 😟

2

1

18

Kelly Marchisio (St. Denis)

@cheeesio

3 months

(3) Challenging tasks degrade fastest. 📉 For example, mathematical reasoning (MGSM) and generative tasks as evaluated by humans and LLM-as-a-Judge suffer a large performance penalty under quantization.

1

2

17

Kelly Marchisio (St. Denis)

@cheeesio

7 months

Have fun! 🤖

Aidan Gomez

@aidangomez

7 months

Here are the weights:

4

19

125

0

2

18

Kelly Marchisio (St. Denis)

@cheeesio

3 years

Our Findings of EMNLP 2021 paper, “An Analysis of Euclidean vs. Graph-Based Framing for BLI from Word Embedding Spaces”, is now public: Code: Paper: *thread* 1/5

An Analysis of Euclidean vs. Graph-Based Framing for Bilingual...

Much recent work in bilingual lexicon induction (BLI) views word embeddings as vectors in Euclidean space. As such, BLI is typically solved by finding a linear transformation that maps embeddings...

arxiv.org

2

17

Kelly Marchisio (St. Denis)

@cheeesio

27 days

Doing my PhD at JHU CS was a true joy! Join them!

Daniel Khashabi 🕊️

@DanielKhashabi

27 days

Computer Science @ JHU is hiring in ALL areas: 🔑 Apply early for flexible scheduling + potential early offer. Our department is expanding fast, especially in AI-adjacent fields. Come join us!

1

17

83

0

2

17

Kelly Marchisio (St. Denis)

@cheeesio

4 years

Finally! Finished my 2019 New Years resolution 🥳🎉☕️ What’s next? I’ve got Hogben’s Mathematics for the Million on the list. (And please excuse the crude coffee mug - a neural network named the color 😅)

2

0

15

Kelly Marchisio (St. Denis)

@cheeesio

2 months

@mayhewsw I expect to update X with a preprint within a few weeks of training completion - stay tuned

1

0

15

Kelly Marchisio (St. Denis)

@cheeesio

4 years

For our 4th date, Martin and I took apart a computer together. For our anniversary, he surprised me with this - the GPU from that night 😭😭😭 "Love you too much to process" -- not quite sure if he's referring to me or the GPU 🤷‍♀️

0

14

Kelly Marchisio (St. Denis)

@cheeesio

1 year

Days ~15-17: Defense date is set! **Wednesday 7 June, 2-4pm** I now have to deliver the full draft to my committee members 2 weeks early, by next Wednesday. I've been sending my advisor draft chapters every few days. Final research content chapter today!

1

0

15

Kelly Marchisio (St. Denis)

@cheeesio

2 years

I've overhauled the ✨IsoVec✨ code to make it more usable -- give it a try! (and feel free to reach out with questions) Github: Paper:

GitHub - kellymarchisio/isovec: IsoVec: Controlling the Relative Isomorphism of Word Embedding...

IsoVec: Controlling the Relative Isomorphism of Word Embedding Spaces (EMNLP 2022) - kellymarchisio/isovec

github.com

Kelly Marchisio (St. Denis)

@cheeesio

2 years

The ability to extract accurate translation dictionaries from monolingual embedding spaces depends critically on their geometric similarity--"degree of isomorphism." We address this root-cause of faulty X-lingual mapping with ✨IsoVec✨ #EMNLP2022 🧵1/N

3

7

49

0

2

15

Kelly Marchisio (St. Denis)

@cheeesio

3 years

in-person @ EMNLP this week - let’s meet for ☕️! (or 🍹🥛🧉🥂🧋🫖🍻🧃!)

0

14

Kelly Marchisio (St. Denis)

@cheeesio

2 years

Day 3: Today I read that Ernest Hemingway allegedly said, “write drunk, edit sober.” Turns out he *didn't* actually say this, which is a real shame because for a moment there I thought he'd cured my writer's block. Anywho, I copied the JHU thesis template today: it exists! 🍻

2

0

14

Kelly Marchisio (St. Denis)

@cheeesio

4 months

A pleasant "good morning" from my dear friend, Command R 🥰

0

13

Kelly Marchisio (St. Denis)

@cheeesio

2 years

Day 2: Skimmed over the thesis of a former lab-mate as an example, then made a rough outline: 3/N

2

0

13

Kelly Marchisio (St. Denis)

@cheeesio

3 months

The ability to serve low-compute models is *critical* for wide global adoption. Even widely-used W8 quantization leads to degradation detectable by humans for some languages, and W4 is even worse. Consider multilinguality as a key evaluation criterion for efficient models!

1

0

12

Kelly Marchisio (St. Denis)

@cheeesio

2 years

Actual footage of me, a vim user, trying to quit nano 🤬

2

0

12

Kelly Marchisio (St. Denis)

@cheeesio

10 months

Come check out our poster tomorrow!

Yihong Chen

@yihong_thu

10 months

If you're deciding which #NeurIPS23 poster to check out tomorrow, don't forget our forgetting paper! Visit poster #328 Thursday morning to dive into the world of active forgetting. Discover how it enhances language models with greater language plasticity. See you there!

2

14

48

0

3

12

Kelly Marchisio (St. Denis)

@cheeesio

3 months

Work by: @cheeesio @TheyCallMeMr_ @hongyucharlie @d_aumiller @ahmetustun89 @sarahookr @seb_ruder 📜

1

0

12

Kelly Marchisio (St. Denis)

@cheeesio

10 months

@ahmetustun89 and I will be chatting about multilingual research at Cohere & C4AI today at #NeurIPS2023 ! Stop by and say “hello”! 👋 🌍🌏🌎

Cohere For AI

@CohereForAI

10 months

@NeurIPSConf @fraser_mince @dzungdinhh This afternoon at 2:30p - 3:30p CT join us at the booth to meet @ahmetustun89 and @cheeesio as they chat with attendees about “Multilingual Research & Innovation at Cohere and Cohere For AI.

0

2

0

6

11

Kelly Marchisio (St. Denis)

@cheeesio

2 years

These are the types of questions & answers that get me excited about using ChatGPT -- The ones that are hard to ask traditional search engines, because punctuation/syntax really matters!

0

1

10

Kelly Marchisio (St. Denis)

@cheeesio

3 years

Same! It me!

Jaclyn A. Siegel, PhD

@jacasiegel

3 years

If you ever see me in person, please say hi. Please approach me at conferences and assume we are best friends. Yes, I want to get coffee or drinks or dinner and talk about your cool new project or hobby or family.

54

226

5K

1

0

9

Kelly Marchisio (St. Denis)

@cheeesio

1 year

Legend. Many a late-night spent watching Professor Strang's lectures on 2x speed to understand my linear algebra homework. The impact this man has had on budding scientists/mathematicians is astounding!

MIT:REGRESSIONS

@mitregressions

1 year

Professor Strang gave his last Linear Algebra lecture today after 66 years at MIT. Strang was among the first to upload his classes to MIT OpenCourseWare when it first came online in the early 2000s. His 18.06 lectures have been viewed millions of times around the world

52

1K

7K

0

9

Kelly Marchisio (St. Denis)

@cheeesio

3 years

Re: Hybrid format -- I know some have felt bogged down with the amount of time it takes to make a recording + (poster / in-person talk) + paper. But, I am "reading" *so* many more of your papers now! I hope if/when we go back to in-person-only, the 10min videos will stay 📖

2

0

9

Kelly Marchisio (St. Denis)

@cheeesio

3 years

Just watched this very clear talk from EMNLP 2021 on Underline. Might help explain our findings in "When Does Unsupervised Machine Translation Work?", particularly Table 5 on instability in BLI ()

arXiv CS-CL

@arxiv_cscl

4 years

Analyzing the Surprising Variability in Word Embedding Stability Across Languages

0

1

2

0

9

Kelly Marchisio (St. Denis)

@cheeesio

4 years

Just received the cutest little “Work From Home Intern” Android from @Google for my remote internship. Thanks to Google Translate Research @markuseful @GrangierDavid for hosting me this summer!

0

9

Kelly Marchisio (St. Denis)

@cheeesio

3 years

@fchollet 22, almost by accident, after a bachelors in psychology/sociology. Changed my life and has brought me more excitement, joy, and fulfillment than I ever could have imagined from a career

0

9

Kelly Marchisio (St. Denis)

@cheeesio

9 months

So excited to work with you, @johnamqdang !

John Dang

@johnamqdang

9 months

Excited to announce that I've joined @cohere as a Research Scholar to work on Multilingual RLHF for LLMs! Thrilled to be working with @ahmetustun89 @cheeesio @KreutzerJulia @sarahookr and the @CohereForAI team!

6

5

105

0

9

Kelly Marchisio (St. Denis)

@cheeesio

10 months

I’ll be at #NeurIPS2023 next week! Let’s meet up!

Sara Hooker

@sarahookr

10 months

Excited to be in New Orleans next week! 🎉 Very proud of the work we will be presenting, with many posters, talks and presentations ahead. Come chat with the @CohereForAI @cohere team. Happy to connect -- looking forward to catching up with friends old and new.

2

9

70

0

8

Kelly Marchisio (St. Denis)

@cheeesio

1 year

@davidbau @iclr_conf @srush_nlp @boknilev I'd love to see a write-up of this after, for those of us who won't be in Rwanda!

0

8

Kelly Marchisio (St. Denis)

@cheeesio

2 years

Day 1: Feeling energized after listening to Episode 151 of @marvettelacy ’s podcast: “Writing a shitty paragraph takes 10 minutes, tops.” Let’s gooooooo! 💪🏽 2/N

2

0

8

Kelly Marchisio (St. Denis)

@cheeesio

1 year

Day 10: Phew! No one tells you (...ok fine, plenty of people told me) that interviewing full-time at the end of a PhD means squeezing in writing in any spare energised moment. 1 hour til liftoff - can I crack out a couple sections? ✈️ ☕️

1

0

8

Kelly Marchisio (St. Denis)

@cheeesio

9 months

@sarahookr We had work about efficient adaptation via the embedding layer alone at ACL and NeurIPS this year!

Jay Alammar

@JayAlammar

1 year

Here's my colleague Kelly Marchisio ( @cheeesio ) presenting “Mini-Model Adaptation: Efficiently Extending Pretrained Models to New Languages via Aligned Shallow Training" Work with @yihong_thu @PSH_Lewis @artetxem at @cohere @forai_ml @ucl_nlp @MetaAI @jhuclsp @RekaAILabs

1

14

60

1

0

8

Kelly Marchisio (St. Denis)

@cheeesio

2 years

Day 4: Printed out my relevant publications, and I'm deciding which parts will be moved to overall intro/background sections vs. which will stay in-chapter with research findings. These two will def need merging, as "BLI for Low Res..." was a follow-on paper to "An Analysis..."

1

0

7

Kelly Marchisio (St. Denis)

@cheeesio

2 years

I’ll be presenting this live in 30mins. Come stop by!

Kelly Marchisio (St. Denis)

@cheeesio

2 years

Might supervised and unsupervised MT be mutually-beneficial? In our #NAACL2022 work, we ask whether the training methods result in systematically different output beyond what is visible via quality metrics like adequacy or BLEU. 🧵1/4

3

5

37

0

1

8

Kelly Marchisio (St. Denis)

@cheeesio

7 months

Coverage of our NeurIPS23 paper, lead by @yihong_thu !

Quanta Magazine

@QuantaMagazine

7 months

To learn more flexibly, a new machine learning model selectively forgets what it already knows. @settostun reports:

0

24

78

0

8

Kelly Marchisio (St. Denis)

@cheeesio

2 years

Day 7: Unexpected 🎁 from my past self: In many of my latex docs, I’d commented-out alternate phrasings, paragraphs that I didn’t have space for, additional derivations, mathematical intuition, etc. Now with unlimited space, these are given new life!

1

0

8

Kelly Marchisio (St. Denis)

@cheeesio

8 months

Alright, hats off to GitHub Copilot 🎩 I wrote only the comments and tiny post-edit to specify behavior of keep, but that's because my prompt was unclear. (OK I know it didn't actually 🖨️, but variable assignment is what I actually wanted so I could play with it myself.)

1

0

8

Kelly Marchisio (St. Denis)

@cheeesio

1 year

Day 11: Personally, I ❤️ the new required "Limitations" section for *ACL conferences. When written well, they clarify work and (counterintuitively?) make the authors' main claims stronger. Keeping them in my thesis!

1

0

7

Kelly Marchisio (St. Denis)

@cheeesio

3 years

Our recent work on bilingual lexicon induction with small parallel corpora is now available online, with code (published at MT Summit 2021) Paper: Code: @jhuclsp

GitHub - kellymarchisio/align-semisup-bli: An Alignment-Based Approach to Semi-Supervised Bilingual...

An Alignment-Based Approach to Semi-Supervised Bilingual Lexicon Induction with Small Parallel Corpora (Marchisio et al, 2021. MT Summit) - kellymarchisio/align-semisup-bli

github.com

0

2

7

Kelly Marchisio (St. Denis)

@cheeesio

2 years

Day 5: Decided that "BLI for Low Res..." and "An Analysis..." definitely belong together under the broader category of *Graph Matching Methods for Bilingual Lexicon Induction*. This morning, I spent an hour combining their setups into a "Shared Experimental Setup" section.

1

0

7

Kelly Marchisio (St. Denis)

@cheeesio

2 years

@FromPhDtoLife Take ~5 years between ugrad & PhD to work, make some money (invest!), have a blast in your early-mid 20s, re-evaluate whether PhD is truly the path for you. If it is, go for it full-force!

0

7

Kelly Marchisio (St. Denis)

@cheeesio

1 year

Day 13: Now that I can talk freely about it, the final chapter is✨Mini-Model Adaptation✨! I got feedback that I should "modernize" my thesis; What does multilinguality from embedding spaces look like in the age of LLMs? Here's a response! #ACL2023NLP

Mini-Model Adaptation: Efficiently Extending Pretrained Models to...

Prior work shows that it is possible to expand pretrained Masked Language Models (MLMs) to new languages by learning a new set of embeddings, while keeping the transformer body frozen. Despite...

arxiv.org

1

0

6

Kelly Marchisio (St. Denis)

@cheeesio

5 years

X-mas gifts received when your BF knows you too well! -- A NN was trained to name new paint colors. This is what it came up with 😂😂

0

6

Kelly Marchisio (St. Denis)

@cheeesio

5 years

Video of my presentation of our recently published work: Found a couple nice summaries of it from MT Summit, also:

Machine Translation Summit 2019 Impressions, Summary and Notes — Part II

This is Part II of my write-up following the recent Machine Translation conference in Dublin.

towardsdatascience.com

WeCNLP

@WeCNLP

5 years

The videos for the invited and lightning talks at WeCNLP 2019 are up! #WeCNLP19

1

5

19

0

6

Kelly Marchisio (St. Denis)

@cheeesio

1 year

Happening in 1hr! Gathertown link here, poster 938:

Cohere For AI

@CohereForAI

1 year

Today we are excited to have work from several of our Cohere Research staff being presented, take a look at where you can find our colleagues @PSH_Lewis @cheeesio @bminixhofer Phil Blunsom @satwik1729 and @tomhosking .

2

6

1

2

6

Kelly Marchisio (St. Denis)

@cheeesio

4 years

There’s a certain joy in reading a book which recommends using a slide rule for division 🥰 ☕️

1

0

6

Kelly Marchisio (St. Denis)

@cheeesio

3 years

Equal Contribution: Kelly Marchisio and Martin St. Denis, 23 Dec 2021. (*some repetition removed)

0

6

Kelly Marchisio (St. Denis)

@cheeesio

2 years

Day 6.5: Wearing this as a critical note-to-self.

2

0

6

Kelly Marchisio (St. Denis)

@cheeesio

1 year

Day 14: Time to re-commit to a writing habit! Interviewing is a full-time job, and each requires prep--so I've fallen off the writing 🚂 recently. To defend in June, I'm re-committing to 1hr writing sessions, 3x/week. Achievable, measurable! 🏆📏 All Aboard!! 🤪🤸🪩

1

0

5

Kelly Marchisio (St. Denis)

@cheeesio

4 months

Excited to be part of this collab! 🗺️

Cohere For AI

@CohereForAI

4 months

📣Announcing our new cross-institutional collaboration. We've brought together researchers invested in improving multilingual benchmarks. We're starting with MMLU, a heavily translated dataset used for multilingual evals that doesn't capture cultural nuances. Let's address this

1

28

114

0

5