Shantanu Chandra

AI Research Scientist@ZS,
MS in AI @ Univ of Amsterdam.

[Shantanu Chandra @ShaanChandra06]

Glad to share that our team came 4th in this Hateful Memes competition among over 3000 participants at #neurips2020 hosted by @facebookai .
Do drop by tomorrow at 5pm PT for the award ceremony and live QnA. !

[AI at Meta @AIatMeta]

Hate speech can come in many forms, including memes that combine text & images. We launched the Hateful Memes Challenge, a first-of-its-kind competition, to help the AI community find new ways to detect multimodal hate speech. Learn about the winners here: https://t.co/I3pOe20sIs

[Shantanu Chandra @ShaanChandra06]

Hi all,
I have summarised the very impressive RetNet paper with some working examples to understand the inner workings in detail. Do let me know your thoughts! :)

#retnet #retentivenetworks #LLMs #Transformers https://t.co/H45xZvPIaz

[Emanuele Rossi @emaros96]

Research in Temporal Graph Learning has been hold back in the past few years by a lack of challenging benchmarks.
I'm super excited to announce Temporal Graph Benchmark (TGB), a collection of seven realistic, large-scale and diverse benchmarks for learning on temporal graphs!

[TheLiverDoc™ @theliverdoc]

I am a Hepatologist.

Arguably, I care for the sickest group of patients in medical practice – end stage liver disease.

I see death almost every day in my ICU.

Some are natural deaths, leading to multiple organ failure triggered by a failing liver.

Some are decided. The family decided to pull the plug. They decide it was time to let go.

And every time, I see the heart rate drop, the blood pressure fall and the electrocardiogram go flat.

And I wonder. What goes on inside a dying brain?
Was it a small black hole, that gets bigger and bigger and then engulfs the field in darkness as the heart shuts down taking the brain away with it?

Or was it just all light at the end of the tunnel and then darkness, once the destination had reached?

A group of doctors did something strange. Something unimaginable.

And it led to them witnessing something melancholic, but beautiful.

They went inside an ICU. Saw that four patients were unconscious.

They were not clinically dead, yet. And not brain-dead. But were comatose.
They were kept alive by a machine that pumped oxygen into their lungs.
Three had brain injury due to cardiac arrest. One had haemorrhaged into the brain.

All were unresponsive for a very long time and sadly, the family decided to pull the plug on them.
To let them go away, peacefully.

But before they did that, the loving family did something bittersweet. They signed a document which was made by the group of doctors, allowing them to study their beloveds’ dying brain.

The group was led by Dr Jimo Borjigin of the Department of Molecular and Integrative Physiology, University of Michigan School of Medicine, Ann Arbor at Michigan.

The doctors fitted machines to the heart and brain of the dying patients and then withdrew their life support by removing the breathing tube.

Immediately after the tube was removed, the brain activity did not show anything different.
But as the brain got closer to death, about 300 seconds before death, a power surge. And it occurred in the regions of the brain which was associated with conscious experience. Like the dying was suddenly yearning to live.

And make no mistake, these were not “some bursts” of electrical activity that happens when the brain cells withered away without oxygen. This was a special kind of activity.

These power surges increased connectivity in a region in the brain called posterior cortical “hot zone” – an area crucial for conscious perception – when it came to just before dying, the brain cells were having a cross-talk, not dying away into oblivion.

They were passing final messages to each other within the regions of the brain, while the physical body lay, limp and motionless.

And the doctors were heartbroken – because the final electrical pattern of surge that the dying brain was showing was the similar to those seen in dreaming humans.

All the patients in the study died.
But before they died, they taught us something.
That even in death, our minds are resilient.
That we clutch at only everything that is good.

What were they dreaming of?
A husband, about his loving wife?
A mother, her children?
Children, their parents?
Or maybe a scene from a movie that changed you?
Or a song under a starry sky?
Or a pet who loved you unconditionally?
We will never know.

But know this, even in death, the mind conquers – with dreams and love, not nightmares and hate.

Remember, nothing perfect lasts forever, except in our memories.

Make good memories while living, so that we can re-live them in death.

Go hug you family, take your children for an ice cream. Walk your dog into the sunset.

Sources:
➡️"Surprising brain activity moments before death" by Dr. F Perry Wilson @fperrywilson via https://t.co/g2NmWlMrAs

➡️Research Article: Surge of neurophysiological coupling and connectivity of gamma oscillations in the dying human brain from @PNASNews https://t.co/YqvZIlDaxN

[Taco Cohen @TacoCohen]

Harm's Law of Smol Models (HLSM) tells us how much we need to scale up the data size (k_D) as we scale down the model size (k_N), if we wish to preserve the loss of a Chinchilla-optimal model.
https://t.co/eOc9pmGGr7

[Ofir Press @OfirPress]

If someone doesn't stop Tim soon he's gonna run Guanaco-65B on a globally distributed cluster of 5 toasters and 3 electric toothbrushes at 70 tokens/sec.

[Dillon Niederhut PhD is at #SciPy2026 @dillonniederhut]

New in @scikit_learn -- experimental support for building models on GPUs with @PyTorch

- @thomasjpfan at #SciPy2023

[Devon ☀️ @devonzuegel]

The saga of Israeli water management is one of the most inspiring stories of the last 75 years

Israel went from having a serious water problem to being a net water exporter to neighboring countries. Meanwhile, the population grew from ~1 million to well over 9 million

[David Pfau @pfau]

We cannot rule out the possibility that radio communication will alert hostile alien superintelligences to our whereabouts, triggering an invasion posing an existential risk to humanity. Therefore we must carefully regulate radio technology until we have better planetary defenses

[Yann LeCun @ylecun]

- Engineer: I invented this new thing. I call it a ballpen 🖊️
- TwitterSphere: OMG, people could write horrible things with it, like misinformation, propaganda, hate speech. Ban it now!
- Writing Doomers: imagine if everyone can get a ballpen. This could destroy society. There should be a law against using ballpen to write hate speech. regulate ballpens now!
- Pencil industry mogul: yeah, ballpens are very dangerous. Unlike pencil writing which is erasable, ballpen writing stays forever. Government should require a license for pen manufacturers.

[Thomas Wolf @Thom_Wolf]

if a paper has appendices you should always read them. always

90% of the time, the interesting pieces you’re looking for are there

[Shantanu Chandra @ShaanChandra06]

@BR365sm When is it gonna come out? :P
Am waiting in IST

[Efstratios Gavves @egavves]

Change the X on arXiv to a 5 and get a beautiful website of the article! So much easier to read! By @JesperDramsch

Fantastic functionality! Maybe the new way of writing papers, so that animations, videos, interactive media is possible?

@NeurIPSConf @CVPR @eccvconf @CSProfKGD

[Shubhendu Trivedi @_onionesque]

A mathematician presents a theorem and proof on the blackboard.

Audience member: That step you described is incorrect <because>, so unfortunately your proof is wrong.

Speaker: No problem, I have another proof.

[Leon Derczynski ⚒️☁️🏔️🌲 @LeonDerczynski]

machine learning researchers learn to optimise their own best paper rate through collusion and other unregulated mechanisms

https://t.co/GsZxab8jCG

[François Chollet @fchollet]

One of the biggest differences I've seen between research and applied ML: in research, most people tune their hyperparameters on the test set to achieve the highest possible score vs. other approaches in the paper's results table

[Phillip Lippe @phillip_lippe]

Happy to finally share our #ICML2022 paper "CITRIS: Causal Identifiability from Temporal Intervened Sequences". We investigate causal representation learning for multidimensional variables, and introduce the concept of "minimal causal variables".
https://t.co/0Rr5eeUHgS
🧵 1/11

[Shantanu Chandra @ShaanChandra06]

Given they submitted it to ICCV unchanged, without acknowledging the original work, shows intent and that they were fully aware of what they were doing.

Help @fahad92sarfraz and team to raise voice against this !

[Shantanu Chandra @ShaanChandra06]

Shocking!
Please raise your voice against this outright plagiarism incident at a top AI conference!
The authors of the said paper were aware of their actions, as their paper was rejected from CVPR on this very basis.
#ICCV #Plagiarism #saynotoplagiarism

[Fahad Sarfraz @fahad92sarfraz]

I'd like to bring attention of DL community to potential plagiarism of our paper and the failure of concerned authorities to address it. (1/2)
https://t.co/JyBrI0tKnS

#research #plagiarism #iccv2021 #ComputerVision #DeepLearning #ArtificialIntelligence

[Phillip Lippe @phillip_lippe]

Are you interested in learning JAX with Flax? We have translated our popular Deep Learning tutorials on CNNs, GNNs, (Vision) Transformers, and more from PyTorch to JAX+Flax, with considerable speedups for smaller models! Check them out here: https://t.co/x38ZWrjZrT
🧵 1/12

[Aran Komatsuzaki @arankomatsuzaki]

Large Language Models are Zero-Shot Reasoners

Simply adding “Let’s think step by step” before each answer increases the accuracy on MultiArith from 17.7% to 78.7% and GSM8K from 10.4% to 40.7% with GPT-3.

https://t.co/ebvxSbac1K