Jyotika Singh (@jyotikasingh_) 's Twitter Profile
Jyotika Singh

@jyotikasingh_

Data Scientist | Python programmer
Interested in NLP, Machine Learning, Deep Learning, Social Media Data, Image, Audio, Speech & Digital Signal Processing

ID: 1162923052222431233

linkhttps://github.com/jsingh811/pyAudioProcessing calendar_today18-08-2019 03:04:07

319 Tweet

263 Followers

99 Following

Jyotika Singh (@jyotikasingh_) 's Twitter Profile Photo

Have something to share with the community #GenerativeAl #Multimodal #Agents? Check out the #CallforPapers @ #GRAILV,#CVPR2026 #LLM #LLMs #genAl #MachineLearning #Artificiallntelligence #DeepLearning #Tranformers #Data #RAG #Vision #VisionLanguage #NLP #naturallanguageprocessing

Jyotika Singh (@jyotikasingh_) 's Twitter Profile Photo

Excited to be co-organizing #GrailV at @cvpr #CVPR2026 Check out the incredible line of speakers and submit your authored work for consideration. #callforpapers #CfP is open. #Conference #Research #Paper #publication #author #genAI #GenerativeAI #LLM #VLM #Multimodal #agents

Jyotika Singh (@jyotikasingh_) 's Twitter Profile Photo

Agents (#Multimodal and #LLM-based) are conversational by design-but we rarely evaluate them that way. A recent large-scale study in 2025 finds that top LLMs perform ~39% worse in multi-turn conversations than in sinhle-turn settings. This surfaces the importance of memory - not

Agents (#Multimodal and #LLM-based) are conversational by design-but we rarely evaluate them that way.

A recent large-scale study in 2025 finds that top LLMs perform ~39% worse in multi-turn conversations than in sinhle-turn settings. This surfaces the importance of memory - not
Jyotika Singh (@jyotikasingh_) 's Twitter Profile Photo

🧑‍💻 **Trying to publish at top venues (NeurIPS / EMNLP / ACL / CVPR / ICML / ICLR)?** After having reviewed 150+ papers and reading over 1000 reviews, here are some tips to avoid low reviewer scores — based on what reviewers *actually* penalize. Most rejects aren’t due to bad

🧑‍💻 **Trying to publish at top venues (NeurIPS / EMNLP / ACL / CVPR / ICML / ICLR)?**

After having reviewed 150+ papers and reading over 1000 reviews, here are some tips to avoid low reviewer scores — based on what reviewers *actually* penalize.

Most rejects aren’t due to bad
Jyotika Singh (@jyotikasingh_) 's Twitter Profile Photo

1️⃣ #PaperReviewStory #1 This one was a first for me. Recently I reviewed a paper that received a low score from a different reviewer mainly because it cited too many #arXiv papers. Reviewer complaint: “Use venue citations where available.” #Takeaway: Even correct citations can

Jyotika Singh (@jyotikasingh_) 's Twitter Profile Photo

2️⃣ #PaperReviewStory #2 Low score because the method was evaluated on only one model. Reviewer comment: “Unclear if this generalizes.” #Takeaway: If it works on one model, reviewers assume it’s fragile — unless you prove otherwise. #TrueStory #AIResearch #PeerReview #MLResearch

Jyotika Singh (@jyotikasingh_) 's Twitter Profile Photo

3️⃣ #PaperReviewStory #3 A reviewer wrote: “Interesting idea, but unclear contribution.” The idea was there — just buried in Section 4. #Takeaway: If your contribution isn’t obvious in 5 minutes, reviewers won’t go hunting for it. #TrueStory #AIResearch #PeerReview #MLResearch

Andrew Ng (@andrewyng) 's Twitter Profile Photo

New course: A2A: The Agent2Agent Protocol, built with @googlecloudtech and IBM Research, and taught by Holt Skinner, Ivan Nardini, and Sandi Besen. Connecting agents built with different frameworks usually requires extensive custom integration. This short course teaches you A2A,

Jyotika Singh (@jyotikasingh_) 's Twitter Profile Photo

4️⃣ #PaperReviewStory #4 One time a reviewer actually tried to implement the provided method in a paper to run an example — and couldn’t do easily with the details shared. Comment implied: “Critical implementation details missing from the paper.” #Takeaway: Double check