Paroma Varma @paroma_varma Twitter profile

Last Seen Profiles

@mundo_jerseys2

@kujionlyfans

@chatham_sdoc

@NomadWarMachine

@PINKO

@solid_run

@TX1TX9

@PolidoriSausage

@ethanchavez256

@VickyMcNally1

@krissastonxxx

@kurukuru_ya

@pIanetMEDIA

@DallasHoll_44

@WillieMcNabb

@Akyro__

@GramyNaMaxa

@JohnsobSte74758

@majidew

@koolselp

@hiambertan

@dyorkside

@ZEturfFr

@stolihaz

@oskar_fm

@KazooDoesArt

@SarahLike007

@CountryDutyZA

@Ncycs

@SirBylHolte

@Ginnewwie

@way0utwest

@vanburen8

@Overdrive

@EdFertility

@honeybearpm

Paroma Varma

@paroma_varma

6 years

Our initial thoughts on debugging training data systematically with ideas from weak supervision and software 2.0! Appreciate any feedback!

0

17

44

Paroma Varma

@paroma_varma

5 years

The Relational Representation Learning Workshop will host a panel with @adityagrover_ @williamleif @jhamrick @thomaskipf @marinkazitnik Submit questions for them here! More information on the website: #NeurIPS2018 #R2L

0

16

33

Paroma Varma

@paroma_varma

5 years

Looking forward to talking about Snuba Tuesday morning at Research Session 1 and at the poster session on Wednesday (poster 1.3) at #VLDB2019 ! Also excited to chat about the Snorkel v0.9 update to programmatically label, transform, and structure training datasets @SnorkelML

Adrian Colyer

@adriancolyer

5 years

"Snuba: automating weak supervision to label training data" Varma & Rè, VLDB'19 #themorningpaper Don't have enough labelled data for your supervised learning project? Snuba can help you to get there by making the very most of what you do have...

3

18

44

0

7

29

Paroma Varma

@paroma_varma

6 years

Papers for #NIPS2018 Relational Representation Learning (R2L) Workshop due 19th Oct! Topics include algorithmic approaches, domain-specific applications, position papers and more: @adityagrover_ @fredsala @zengola @ProfJenNeville @ermonste @HazyResearch

1

11

27

Paroma Varma

@paroma_varma

5 years

Using #snorkelML over real-world @ukbiobank heart MRI data to find individuals with dysfunctional aortic valves (~1% of population). Key: programmatically labeling unlabeled data to create a representative training set!

Nature Communications

@NatureComms

5 years

@jasonafries and colleagues provide a deep learning model for aortic valve malformation classification using unlabeled cardiac MRI sequences. @paroma_varma @vincentsunnchen @jdunnmon @MFiterau @MobilizeCenter @HazyResearch @euanashley @JamesRPriestMD

0

15

22

0

12

19

Paroma Varma

@paroma_varma

6 years

Debugging #ML pipelines was a hot topic at the recent DAWN retreat! Shaping training data, exploiting log data, and model introspection were some of the popular themes that came up in conversation with our industry affiliates

0

12

19

Paroma Varma

@paroma_varma

5 years

Some thoughts on weak supervision, our system #Snorkel , and vision for *massively multitask* weak supervision systems!

Stanford AI Lab

@StanfordAILab

5 years

Read our newest blog post! On using weak supervision, or high-level noisy sources of labels, to efficiently label training data. Courtesy of @paroma_varma , @ajratner , @bradenjhancock , and others at Professor Chris Ré's @HazyResearch lab.

1

58

138

0

3

16

Paroma Varma

@paroma_varma

5 years

SOTA on SuperGLUE with Snorkel! Programmatically building training datasets by 1) labeling 2) transforming and 3) partitioning as the key to success!

vincent sunn chen

@vincentsunnchen

5 years

We have SOTA on the SuperGLUE benchmark! Our experiment: how far can we get with a basic model + training data operations? Key abstractions for programming your training data: 1) labeling 2) augmentation 3) partitioning

3

39

117

0

3

12

Paroma Varma

@paroma_varma

6 years

Making it easier to debug machine learning models and data it learns from - excited to try to it!

Google AI

@GoogleAI

6 years

Today, we are launching the What-If Tool, a new feature of the open-source TensorBoard web application, which let users analyze and better understand an ML model without writing code. We look forward to people using, and contributing to, the What-If Tool.

16

538

1K

0

9

Paroma Varma

@paroma_varma

5 years

Check out our work on learning graph structure without ground truth labels! Incorporating this structure into weak supervision models leads to high-quality training labels 🚀

Fred Sala

@fredsala

5 years

Excited to talk about our work on learning graph structures for Snorkel & weak supervision at #icml2019 ! Stop by our talk at 5:10pm in Room 104 and our poster #119 at 6:30pm in the Pacific Ballroom today! Blog: Paper:

0

5

20

0

4

7

Paroma Varma

@paroma_varma

5 years

Awesome tweet summary of @sppalkia ’s talk on Weld, a JIT compiler/IR in Rust! Work with @matei_zaharia @deepakn94 @jamesjthomas @parimarjan @rahulpalamut and Pratiksha Thaker

0

1

5

Paroma Varma

@paroma_varma

5 years

@vincentsunnchen @JeffDean Awww thank you @vincentsunnchen !! Working with you has been amazing and I’ve learned so much from you!🎉

0

4

Paroma Varma

@paroma_varma

4 years

@BerkeleyML @SnorkelAI Thanks for having me! Really enjoyed the discussions!

0

4

Paroma Varma

@paroma_varma

5 years

@HazyResearch @JeffDean @ajratner @bradenjhancock Some of our work on #Snorkel and vision for massively multitask weak supervision systems here!

0

3

Paroma Varma

@paroma_varma

6 years

Super helpful “talk first” method to design a poster! Much better than simply summarizing the paper

Xanda Schofield (she/her)

@XandaSchofield

6 years

Unsure of how to start designing a conference poster? Frustrated that your research posters often end up too verbose, or full of content you don't talk about? @dmimno and I wrote up the "talk-first" poster design workflow used by our research group!

2

41

117

0

3

Paroma Varma

@paroma_varma

5 years

@adityagrover_ @fredsala @zengola @ProfJenNeville @ermonste @HazyResearch Submit questions for panelists here!

0

1

Paroma Varma

@paroma_varma

6 years

@jamesjthomas wow!

0

1