Harsh Shrivastava

Research interest in Machine Learning. I like poetry. Always curious :)

[Harsh Shrivastava @Ace_Harsh]

Today's achievement! #2048

[J. AI Research-JAIR @JAIR_Editor]

New Article: "Methods for Recovering Conditional Independence Graphs: A Survey" by Shrivastava and Chajewska https://t.co/cP4LO51YIL

[Harsh Shrivastava @Ace_Harsh]

@windowsreport Thank you for the wonderful summarization of this work!

[Windows Report @windowsreport]

Microsoft researchers discovered Generative Kaleidoscopic Networks, and it can drastically enhance AI’s creativity.

https://t.co/cguXoRawx8

[Harsh Shrivastava @Ace_Harsh]

Kindly reach out with your feedback and possible ways to explore this #GenerativeAI based research further!

[Harsh Shrivastava @Ace_Harsh]

Thank you @MSFTResearch for highlighting our work on "Generative Kaleidoscopic Networks"! I was trying to create a dataset kaleidoscope using neural networks. In this process, I additionally discovered that neural networks demonstrate an "over-generalization" phenomenon. (1/2)

[Microsoft Research @MSFTResearch]

In this issue: Generative kaleidoscopic networks; Text diffusion with reinforced conditioning; PRISE – Learning temporal action abstractions as a sequence compression problem: https://t.co/7ZzU8eE3xw

[Harsh Shrivastava @Ace_Harsh]

Tldr (algorithm): Akin to the fractal generation process, repeatedly apply learned deep ReLU networks on an input noise to generate samples.

Paper & Code: https://t.co/CWzmk3RnQv

Celeb-A dataset Kaleidoscope trained on a subset. (2/2)

[Harsh Shrivastava @Ace_Harsh]

Very interesting work, Gaurav @datacrunch3r ! Given the compute efficiency, it will be great to see these models deployed on Mobile phones!

[Gaurav Shrivastava @datacrunch3r]

I am pleased to announce that our work Video Decomposition Prior(VDP) have been accepted for ICLR 2024.

Key Highlights:

(1/3)

[Harsh Shrivastava @Ace_Harsh]

Thank you @MSFTResearch for sharing this work! Kindly reach out to me (or my co-author) in case you want to know more or explore potential applications! Thanks :)

[Microsoft Research @MSFTResearch]

Neural Graphical Models (NGMs) provide a solution to the challenges posed by traditional graphical models, offering greater flexibility, broader applicability, and improved performance in various domains. Learn more: https://t.co/MUYYZGQWN0

[Microsoft Research @MSFTResearch]

Neural Graphical Models (NGMs) provide a solution to the challenges posed by traditional graphical models, offering greater flexibility, broader applicability, and improved performance in various domains. Learn more: https://t.co/MUYYZGQWN0

[Harsh Shrivastava @Ace_Harsh]

We found that some regularization terms perform much better than the basic updates. We are working on a theory to explain their effectiveness of doing KP over CI graphs. Please reach out if you're interested to know more!

[fly51fly @fly51fly]

This paper introduces an algorithm for knowledge propagation on conditional independence graphs, improving analysis in various domains.

[Harsh Shrivastava @Ace_Harsh]

I joke about it now.. 😇 but realizing thought-controlled computing will be cool!

[antisense. @razoralign]

Reconstruction of Gene Regulatory Networks using sparse graph recovery models https://t.co/aDnIAbVIy4 Reconstruction of Gene Regulatory Networks using sparse graph recovery models https://t.co/aDnIAbVIy4

[Harsh Shrivastava @Ace_Harsh]

Excited to release our framework for doing multivariate time series segmentation in O(N). It is a cross domain approach based on using sparse graph recovery methods. Do check it out!

[cs.LG Papers @arxiv_cs_LG]

Are uGLAD? Time will tell!. Shima Imani and Harsh Shrivastava https://t.co/T8RxjkJoB6

[cs.LG Papers @arxiv_cs_LG]

Are uGLAD? Time will tell!. Shima Imani and Harsh Shrivastava https://t.co/T8RxjkJoB6

[DAIR.AI @dair_ai]

Top ML Papers of the Week (Mar 6 - Mar 12):

- Visual ChatGPT
- PaLM-E
- A History of Generative AI
- MathPrompter
- Foundation Models for Decision Making
- GigaGAN
...

[Harsh Shrivastava @Ace_Harsh]

@omarsar0 Thank you for your interest in our work! Yes, we are actively exploring ways to expand this approach for other problems :)

[AK @_akhaliq]

MathPrompter: Mathematical Reasoning using Large Language Models

improves over state-of-the-art on the MultiArith dataset (78.7% → 92.5%) evaluated using 175B parameter GPT-based LLM

abs: https://t.co/A2N09LLQsX

[Harsh Shrivastava @Ace_Harsh]

@omarsar0 Thank you for sharing our work :)

[elvis @omarsar0]

MathPrompter: a technique that improves LLM performance on mathematical reasoning problems.

It uses zero-shot chain-of-thought prompting and verification to ensure generated answers are accurate.

https://t.co/vjJ0MJBZDC

[elvis @omarsar0]

MathPrompter: a technique that improves LLM performance on mathematical reasoning problems.

It uses zero-shot chain-of-thought prompting and verification to ensure generated answers are accurate.

https://t.co/vjJ0MJBZDC