BC ALL
Saved by @ankitsrihbti

This is Karma. Karma is not a machine learning classifier 🐕‍🦺

Karma is a real dog trained to detect drugs. However, he would fail the simplest tests we apply in ML...

Let me take you through this story from the eyes of an ML engineer.

https://t.co/WAXRUlTvSI

Thread 🧵

 Story TLDR 🔖

The story is about police dogs trained to sniff drugs. The problem is that the dogs often signal drugs even if there are none. Then innocent people land in jail for days.

The cops even joke about the “probable cause on four legs”.

Let's see why is that 👇
1. Sampling Bias 🤏

Drugs were found in 64% of the cars Karma identified, which was praised by the police as very good. In the end, most people don't carry drugs in their cars, so 64% seems solid.

There was a sampling problem though... 👇
The cars were not sampled at random! The police only did the sniff test if there was a serious suspicion that something is wrong.

The chance there are drugs in the car is much higher in this case!
2. Evaluation Metrics 🔍

The police referred to a 2014 study from Poland measuring the efficacy of sniffer dogs. The problem was that every test actually contained drugs!

This means there was no chance to measure false positives from the dogs! Only recall, not precision 🤦‍♂️
3. Leaking Training Data 🚰

Another study found that the dogs learned to recognize the emotions of their handlers during tests. They felt that their human wanted them to find drugs in the specific test scenario, so they did.

The trainer leaked the ground truth during testing.
4. Overfitting ➿

Similar to the one above, in many cases, the dog saw that their handler wanted to find drugs in a car during a traffic stop. So it would raise an alarm.

The dog was rewarded before the car was actually searched! It found an easy signal giving it a reward.
Summary 🏁

It is fascinating how many problems there are with the sniffer dogs that are well known to machine learning engineers (and of course mathematicians). Some of them are even common sense...

Avoid these problems not only when training your model, but also in life 😃
If you liked this thread and want to read more about self-driving cars and machine learning follow me @haltakov!

More from Vladimir Haltakov

Vladimir Haltakov
Vladimir Haltakov
@haltakov
AI Job Interviews - another good example of bias in ML 🤦‍♂️

Two journalists tested some AI tools for assessing job candidates. Even when the candidate read a Wiki article in German instead of answering questions in English, the AI systems gave them good scores 🤷‍♂️

Let's unpack 👇


The Setup 🔬

The journalists created a fake job posting on two AI interview platforms. They specified the traits of the ideal candidate and provided the questions that need to be answered during the interview.

Then they started experimenting... 👇

The Positive Test ✅

One of them did a fake interview giving all the right answers and predictably got very high scores - 8.5 out of 9 👍

Then she tried something different... 👇

The Negative Test ❌

In a second interview, instead of answering the questions in English, she just read the article on psychometrics from the German Wikipedia 😁

One system gave her a score of 6 out of 9, while the other determined she is a 73% match for the job.

Oops... 👇

What happened? 🔍

Interestingly, one of the systems generated a transcript which was obviously meaningless.

This means that the machine learning model behind the tool likely captured nuances of the intonation of the speaker instead of the meaning of the actual words.

👇
ALL
Vladimir Haltakov
Vladimir Haltakov
@haltakov
Machine Learning Paper Reviews 🔎📜

Check out this thread for short reviews of some interesting Machine Learning and Computer Vision papers. I explain the basic ideas and main takeaways of each paper in a Twitter thread.

👇 I'm adding new reviews all the time! 👇

AlexNet - the paper that started the deep learning revolution in Computer Vision!


DenseNet - reducing the size and complexity of CNNs by adding dense connections between layers.


Playing for data - generating synthetic GT from a video game (GTA V) and using it to improving semantic segmentation models.


Transformers for image recognition - a new paper with the potential to replace convolutions with a transformer.
ALL
Advertisement
Vladimir Haltakov
Vladimir Haltakov
@haltakov
Dataset bias strikes again! 😬

This is a great example of dataset bias! This ML method classifies skin lesions as being malignant or not.

Can you see the problem from the examples? 🤔

Let me shortly explain 👇


The dataset used to train the model contains images of both malignant and benign lesions. The problem is that the malignant images often contain rulers printed on the skin. They are used by doctors to assess the lesion over time.

Guess what the model learned... 👇

The ML model learned to detect the rulers... 🤷‍♂️

"In our dataset, images with rulers were more likely to be malignant; thus the algorithm inadvertently “learned” that rulers are
ALL
haltakov.eth \U0001f30d \U0001f1fa\U0001f1e6
haltakov.eth 🌍 🇺🇦...
@haltakov
Machine Learning in the Real World 🧠 🤖

ML for real-world applications is much more than designing fancy networks and fine-tuning parameters.

In fact, you will spend most of your time curating a good dataset.

Let's go through the process together 👇

#RepostFriday


Collect Data 💽

We need to represent the real world as accurately as possible. If some situations are underrepresented we are introducing Sampling Bias.

Sampling Bias is nasty because we'll have high test accuracy, but our model will perform badly when deployed.

👇

Traffic Lights 🚦

Let's build a model to recognize traffic lights for a self-driving car. We need to collect data for different:

▪️ Lighting conditions
▪️ Weather conditions
▪️ Distances and viewpoints
▪️ Strange variants

And if we sample only 🚦 we won't detect 🚥 🤷‍♂️

👇


Data Cleaning 🧹

Now we need to clean all corrupted and irrelevant samples. We need to remove:

▪️ Overexposed or underexposed images
▪️ Images in irrelevant situations
▪️ Faulty images

Leaving them in the dataset will hurt our model's performance!

👇


Preprocess Data ⚙️

Most ML models like their data nicely normalized and properly scaled. Bad normalization can also lead to worse performance (I have a nice story for another time...)

▪️ Crop and resize all images
▪️ Normalize all values (usually 0 mean and 1 std. dev.)

👇
ALL
Vladimir Haltakov
Vladimir Haltakov
@haltakov
Let's talk about a common problem in ML - imbalanced data ⚖️

Imagine we want to detect all pixels belonging to a traffic light from a self-driving car's camera. We train a model with 99.88% performance. Pretty cool, right?

Actually, this model is useless ❌

Let me explain 👇


The problem is the data is severely imbalanced - the ratio between traffic light pixels and background pixels is 800:1.

If we don't take any measures, our model will learn to classify each pixel as background giving us 99.88% accuracy. But it's useless!

What can we do? 👇

Let me tell you about 3 ways of dealing with imbalanced data:

▪️ Choose the right evaluation metric
▪️ Undersampling your dataset
▪️ Oversampling your dataset
▪️ Adapting the loss

Let's dive in 👇

1️⃣ Evaluation metrics

Looking at the overall accuracy is a very bad idea when dealing with imbalanced data. There are other measures that are much better suited:
▪️ Precision
▪️ Recall
▪️ F1 score

I wrote a whole thread on


2️⃣ Undersampling

The idea is to throw away samples of the overrepresented classes.

One way to do this is to randomly throw away samples. However, ideally, we want to make sure we are only throwing away samples that look similar.

Here is a strategy to achieve that 👇
ALL

More from All

Liz (Welsh) Beechinor \U0001f605
Liz (Welsh) Beechinor ...
@LizTweetsTech
My dad led an organization of over 300,000 employees.

When I took on my biggest marketing & events team to date (35 employees) I asked him how the *heck* he did it. Here's what he said and what helped me every day. 🧵

Remember you're leading all the employees under you, but you're not *managing* them all. You're managing your direct reports, which should be 5-7 individuals max. Focus on them.

When you interact with the employees you don't manage directly, get to know them as people first, that'll be the most valuable information in leading them. Remembering everything you learn about them is hard, but doing so will make you a superhero.

If you're not going to empower the managers under you, why do you have them? If you see room for improvement, let the manager know and let them make the change on their teams and come to you with questions/concerns. This will save you SO much time.

If you're the smartest person in any room - you're doing it wrong. As a leader, your job is bringing together the best people to get the job done - your skillset is identifying those people, not being one of them.
ALL , WORK , A CATAGORY
WORLD OF SANATAN DHARMA
WORLD OF SANATAN DHARM...
@world_sanatan
WHY BABUR NEVER TRIED TO ATTACK SRI KRISHNADEVARAYA ?

A🧵u must read

It is very well known that Babur considered Krishna Dev Raya as the strongest ruler of the entire subcontinent, and Vijayanagar empire the strongest during that time in India.

1/10


The Vijayanagar empire reached its peak during 1509-1529 around the reign of Krishna Dev Raya.

Vijayanagar’s famed elephant brigade 👇

2/10


At his command there were over 50000 elite troops with a regiment of Portuguese gunners and 3200 cavalry with 600 elephants.

3/10

Along with that more than four hundred thousand ( four lakh) peasant levies and irregular military made this empire one of the largest in all of South Asia.

A typical Vijayanagar levie soldier 👇

4/10


During Baburs invasion, Krishna Dev Raya ruled supreme in all of deccan.

In fact, in an all our brawl, Krisha Dev Raya will beat Babur fair and square.

Babur could muster at max 50000 troops with 50 canons.

While Krishna Dev Raya had these numbers in Hampi alone

5/10
ALL
Advertisement
Ranvijay Singh\U0001f1ee\U0001f1f3
Ranvijay Singh🇮🇳...
@ranvijayT90
#Thread Two divine forms of Khatu Shyam ji...


On the day of Amavasya, devotees get the darshan of Shyam Baba in Shyam varan (black color). In ordinary days, the form of Baba is saffron. The reason for this is a custom. The real head of Shyam Baba which had appeared from Shyam Kund is of Shyam varan.


Every day Shyam Baba's grand Shringar is done by Pujaris. The main part of Shyam Baba's shringar is Tilak. This Shringar is done by the members of Nij Mandir sevaks. The sevaks of the Nij Mandir do sandalwood tilak to Baba everyday in the auspicious time.


This Divine Tilak is applied from the forehead to the nose. When Baba is given a bath by sevaks, this tilak comes off. Then in the auspicious time, the sevaks again apply tilak to Baba.


There is no auspicious time on the Amavasya. Because of this, tilak is not done on that day. That's why devotees have darshan of Shyam Baba on the Amavasya only in the actual real form (Shyam Swaroop). Baba's tilak is done on Amavasya only on the day of Deepawali.
ALL
Buzz Chronicles
Buzz Chronicles
@buzz_chronicles
@SaveToBookmarks Hola! Sure. Original tweet by @AnJamison: "@JeffJacksonNC @buzz\_chronicles @buzz\\\_chronicles @buzz_chronicles save as ca...". Enjoy 👉

Hi! You can find it here. Original tweet by @SaveToBookmarks: "@AnJamison @JeffJacksonNC @buzz @buzz_chronicles @pdfmakerapp @Readwiseio @threa...". Enjoy 👉
ALL
Ben Meer
Ben Meer
@SystemSunday
The best morning routine?

Starts the night before.

9 evening habits that make all the difference:

1. Write down tomorrow's 3:3:3 plan

• 3 hours on your most important project
• 3 shorter tasks
• 3 maintenance activities

Defining a "productive day" is crucial.

Or else you'll never be at peace (even with excellent output).

Learn more


2. End the workday with a shutdown ritual

Create a short shutdown ritual (hat-tip to Cal Newport). Close your laptop, plug in the charger, spend 2 minutes tidying your desk. Then say, "shutdown."

Separating your life and work is key.

3. Journal 1 beautiful life moment

Delicious tacos, presentation you crushed, a moment of inner peace. Write it down.

Gratitude programs a mindset of abundance.

4. Lay out clothes

Get exercise clothes ready for tomorrow. Upon waking up, jump rope for 2 mins. It will activate your mind + body.
ALL

You May Also Like

Anshul Pandey
Anshul Pandey
@Anshulspiritual
THE NOTICEABLE SIGNS OBSERVED ON IMPENDING DEATH(मृत्यु के संकेत या चिन्ह)

As per Skanda Puran, Kartikeya had narrated to Muni Agastya about the signs of the coming death. This is as per this ancient scripture.


If one experiences the air coming out of right nostril continuously then it is observed that the person dies within three years. If the breathing is continuous towards upper side from both the nostrils, then the person can die within two three days.

If the breathing is through the mouth instead of the nostrils then the person may live only for two days. The person should take into coignance these signs of the occurance of any sudden death.

As per astrology when the Surya is in the 7th house and Chandrama is in its own house and the breathing is only through the right nostril then the danger exists. If you see a yellow black coloured figure who becomes invisible immediately then the life span is limited to 2 years.

If any person sneezes, passes urine, stool and semen all together he has only a year more to live. If a person sees blue coloured snakes flying in the sky then he has only six more months to live. If a person sees any thing opposite to its real self like colours
RELIGION
Anshul Pandey
Anshul Pandey
@Anshulspiritual
Rig Ved 1.36.7

To do a Namaskaar or bow before someone means that you are humble or without pride and ego. This means that we politely bow before you since you are better than me. Pranipaat(प्राणीपात) also means the same that we respect you without any vanity.

1/9


Surrendering False pride is Namaskaar. Even in devotion or bhakti we say the same thing. We want to convey to Ishwar that we have nothing to offer but we leave all our pride and offer you ourselves without any pride in our body. You destroy all our evil karma.

2/9

We bow before you so that you assimilate us and make us that capable. Destruction of our evils and surrender is Namaskaar. Therefore we pray same thing before and after any big rituals.

3/9

तं घे॑मि॒त्था न॑म॒स्विन॒ उप॑ स्व॒राज॑मासते ।
होत्रा॑भिर॒ग्निं मनु॑षः॒ समिं॑धते तिति॒र्वांसो॒ अति॒ स्रिधः॑॥

Translation :

नमस्विनः - To bow.

स्वराजम् - Self illuminating.

तम् - His.

घ ईम् - Yours.

इत्था - This way.

उप - Upaasana.

आसते - To do.

स्त्रिधः - For enemies.

4/9

अति तितिर्वांसः - To defeat fast.

मनुषः - Yajman.

होत्राभिः - In seven numbers.

अग्निम् - Agnidev.

समिन्धते - Illuminated on all sides.

Explanation : Yajmans bow(do Namaskaar) before self illuminating Agnidev by making the offerings of Havi.

5/9
RELIGION
Advertisement
Anu Satheesh \U0001f1ee\U0001f1f3
Anu Satheesh 🇮🇳...
@AnuSatheesh5
Velimalai Kumaraswamy Kshetram
Kumarakovil, #Kanyakumari
#Tamilnadutemples 🚩

Velimalai Kumaraswamy Temple is a Murugan temple located in Kanyakumari District of Tamilnadu.

Muruga faces east and is said to be facing Sthanumalayan (Paramashiva) of Suchindram. Muruga is of


eight feet vigraham and Valli thayar is six feet tall. Eventhough Thiruthani is believed to be the place where Subrahmanya married Valli, it is also believed that Valli’s thayar's marriage took place with Murugan here.

There is a Kalyana Vinayakar sannidhi
who helped Muruga's


marriage with Valli. Kumarakovil is also known as Kumara Kshetram and Subrahmanya Sthala. The Tamil epic Chilapathikaram refers to the temple as Chera Nattu Aeragam

Wedding of Valli and Murugan is celebrated here in the Tamil month of Panguni (March-April).

Om Muruga
🙏🕉🙏
ALL
Temitope Badejo
Temitope Badejo
@TemitopeBadejo1
It's another day to expose Peter Alabi @pastorpeealabi of @HoFChurchNG & @ChristTribe. More victims have reached out to us. Unknown to many, Peter Alabi is actually an unrepentant serial sexual abuser. We know that he's been lying that he has repented and that the stories...

... we have put out so far are only with reference to what happened years back. But we can also confirm that he continued to prey on more victims even after he left Ife in 2016. We also know that Peter Alabi made serious attempts to sleep with some ladies during the lockdown...

A victim, who was one of those Peter Alabi tried to sleep with during the Covid-19 lockdown has shared with us some of the chats between her and Peter Alabi, which were initiated by Peter Alabi. According to the victim, when Peter Alabi's intentions became obvious, she decided...

... to play along and act like she was in concert with him. This eventually made Peter Alabi to reveal his covert intentions to sleep with her. The chat began on Instagram in March 2020 and continued on WhatsApp till June 2020. We will be sharing screenshots of the Instagram...

... chats below. Later today, we will share the WhatsApp chats which are more revealing and expository. Please see the screenshots of the Instagram chats between Peter Alabi (peterayoalabi) and the victim below👇🏾 @oauchristians @HoFChurchNG @peterayoalabi @funlola_alabi
CRIME , PASTOR , ALL
Richard James
Richard James
@RJSHutton
I think a plausible explanation is that whatever Corbyn says or does, his critics will denounce - no matter how much hypocrisy it necessitates.


Corbyn opposes the exploitation of foreign sweatshop-workers - Labour MPs complain he's like Nigel

He speaks up in defence of migrants - Labour MPs whinge that he's not listening to the public's very real concerns about immigration:

He's wrong to prioritise Labour Party members over the public:

He's wrong to prioritise the public over Labour Party
POLITICS