kasrasadatsharifi

[kasrasadatsharifi @kasrasa]

Can modern CNNs still compete with Vision Transformers?
In my latest article, I explore ConvNeXt’s architecture, its trade-offs, and how it performs in practical experiments.
Read it on @Towards_AI
https://t.co/1ZW5QzxucO
#DeepLearning #ComputerVision #ConvNeXt #MachineLearning

[kasrasadatsharifi @kasrasa]

2/2
The review fixed 66 mistakes but changed 28 correct predictions into wrong ones: a net gain of 38, at about 1.3 seconds per review.
Whether that trade-off makes sense depends on the application.
I explain the experiment in my @towards_AI article:
https://t.co/Xj66vi8dTv

[kasrasadatsharifi @kasrasa]

1/2
In production, images often outnumber labels. I wanted to know what a VLM could add when a strong classifier was available.

On a small Food-101 subset, ConvNeXt performed better than SmolVLM on its own. So I tried using SmolVLM to review ConvNeXt’s uncertain predictions.

[kasrasadatsharifi @kasrasa]

2/2
Adapting the model’s first layers to 32×32 images made the biggest difference: the k-nearest-neighbor score rose from about 87% to 93%. I walk through the experiments in @LevelUpCoding:
https://t.co/Jd57MclRkq
#ResNet #CIFAR10 #MachineLearning #ComputerVision

[kasrasadatsharifi @kasrasa]

1/2
I wanted to see how much a ResNet’s training setup matters on CIFAR-10. I started with a ResNet-18 trained from scratch and measured both classification and how useful its learned features were for k-nearest-neighbor classification.

[kasrasadatsharifi @kasrasa]

2/2
Fine-tuning DeiT’s last five blocks beat full fine-tuning in this run. ConvNeXt had the highest accuracy, while calibration gave me another reason to look beyond that single score.

I wrote about it in my @LevelUpCoding_ article:
https://t.co/8XCYMUwmAM

[kasrasadatsharifi @kasrasa]

1/2
When labeled data and compute are limited, how much of a pretrained vision model should you fine-tune?

I tested several approaches with DeiT-Tiny on a small Food-101 subset, then compared it with ResNet-50 and ConvNeXt-Tiny.

[Oli London @OliLondonTV]

Iranians hold memorial service for teenage athlete Arnika Debagh.

She was a talented swimming champion who had won multiple national titles in Iran.

She died after being shot in the heart by Iranian security forces.

💔

[Alborz @Elllen88]

It’s time to prioritize Iranians who came legally, studied, worked, applied for PR, and have been waiting due to prolonged security checks. They are contributing to Canada’s economy. Resources should not be wasted on fraudulent asylum claims.
#DelayedIranianApplications

[Alborz @Elllen88]

It’s time to prioritize Iranians who came legally, studied, worked, applied for PR, and have been waiting due to prolonged security checks. They are contributing to Canada’s economy. Resources should not be wasted on fraudulent asylum claims.
#DelayedIranianApplications

[Alborz @Elllen88]

@CitImmCanada It makes sense to hold government officials accountable, not ordinary law-abiding Iranians. We're asking for a fair process that doesn't discriminate but still ensures Canada's safety. Let's not conflate citizens with the actions of their government.

#DelayedIranianApplications

[Alborz @Elllen88]

@MarcMillerVM It’s time to prioritize Iranians who came legally, studied, worked, applied for PR, and have been waiting due to long security checks. They are contributing to Canada’s economy. Resources should not be wasted on fraudulent asylum claims.
@CitImmCanada
#DelayedIranianApplications

[Hamed Maleki @hmalekid]

Call for Action: Ensure Fair Processing for Permanent Residency Applicants https://t.co/o7xkzE1QC4 @MarcMillerVM @shafqatalimp @sukhdhaliwal @BradRedekopp @F_ElKhoury @Alduceppe @BardishKW @MichaelParsa @HonAhmedHussen #DelayedIranianApplications #FairnessForIranians

[Hamed Maleki @hmalekid]

Iranian applicants face PR suspensions due to security checks. We urge you to back this petition: https://t.co/o7xkzE1QC4 @shafqatalimp @sukhdhaliwal @BradRedekopp @F_ElKhoury @Alduceppe @BardishKW @MichaelParsa @HonAhmedHussen #DelayedIranianApplications

[Hamed Maleki @hmalekid]

We need transparency and fairness for Iranian applicants. Call for Action: Ensure Fair Processing for Permanent Residency Applicants https://t.co/o7xkzE2orC @MarcMillerVM #DelayedIranianApplications

[Hamed Maleki @hmalekid]

A landmark Federal Court ruling sets a precedent against prolonged security delays. Details: [https://t.co/1cjzRN1RlG] @MarcMillerVM #DelayedIranianApplications

[Hamed Maleki @hmalekid]

We're a group of Iranian applicants stuck in our PR applications' security step.A landmark Federal Court ruling sets a precedent against prolonged security delays. [https://t.co/1cjzRN1RlG] It's time for action: https://t.co/o7xkzE1QC4 @MarcMillerVM #DelayedIranianApplications

[Hamed Maleki @hmalekid]

Iranian applicants face PR suspensions due to security checks. A 2022 NSIRA investigation confirmed Iranians are systematically flagged. We urge @MarcMillerVM to back this petition: https://t.co/o7xkzE1QC4 #DelayedIranianApplications

[Hamed Maleki @hmalekid]

It's concerning that @CanBorder/@csiscanada send law-abiding Iranians to enhanced security checks while fast-tracking people with questionable histories. We urge you to back us in a faster, fairer process—without compromising Canada's safety. #DelayedIranianApplications

[Hamed Maleki @hmalekid]

The security check for my PR application has taken months with no end in sight.This delay is affecting my ability to work and support my family.We ask for timely & fair processing of our applications. Please sign & share this: https://t.co/o7xkzE1QC4 #DelayedIranianApplications