Microsoft

Started as a Data Scientist and grew to Principal Data Scientist Manager.
My team has supported multiple industries such as retail, gaming, finance, marketing, e-commerce, energy, media, manufacturing, and others. We implement solutions in a wide range of AI areas including recommendation systems, natural language processing, computer vision, time series forecasting, and other machine learning techniques.
Our repository Microsoft Recommenders ➡️➡️ https://github.com/microsoft/recommenders is a 
world reference in recommendation systems. Starred and used by +10k people.

Principal Data Scientist Manager

IE University

I'm helping a new generation of students to understand and apply AI.

Adjunct Professor

Imperial College London

I help the College to define its strategic development, research themes, academic programs, and partnerships with the industry.

Industry Board Member at Dyson School of Design Engineering

samsamia

We developed a SaaS solution to increase fashion e-commerce revenue through visual search. Our technology allowed users to take a picture of a garment they liked, and find the most similar garment within the customer's catalog. I led the engineering and sales team. Among our customers, we had one of the biggest e-commerce in the UK and one of the biggest media companies in Spain.

CEO & Co-Founder

King's College London

Research topics: Learning from Demonstration, Human-Robot Interaction, Markov Chains, and 
Non-linear control.

External Researcher

Universidad Carlos III de Madrid

Research in Machine Learning for robotics: Learning from Demonstration, Computer Vision, kinematic and dynamic control of humanoid robots, and non-linear control. See my publications: ➡️➡️ https://github.com/miguelgfierro/publications

Professor at several Engineering degrees teaching Control Engineering, Industrial Automation, and Industrial Robotics.

Research Assistant at Robotics Lab

Development of algorithms to predict the energy price fluctuation in Spanish electrical market focusing in wind turbines.

Scholarship Researcher in Electric Department

Asociación de Robótica de la Universidad Carlos III

A non-profit student organization at Universidad Carlos III de Madrid. Development of R&D robotics projects including mobile robots, aerial robots (Unmanned Aerial Vehicles UAVs), 3D printers, and small humanoid robots.

President & Co-Founder

I lead the Personalization team at Microsoft working on Reco Systems, NLP, Computer Vision, and other ML solutions.

In addition, I help people understand and apply AI. Whether you want to switch your career to Data Science, land a job in a big tech company, grow your Data Science career, or apply AI to your business, I can help you.

Subscribe to my list ➡️➡️ https://bit.ly/emaillistmiguel ⬅️⬅️

____________________________
Blog: https://bit.ly/miguelgfierro
GitHub: https://github.com/miguelgfierro
Microsoft Recommenders: https://github.com/microsoft/recommenders
Publications and conferences: https://github.com/miguelgfierro/publications

I am at my maximum connections, please follow me and press the bell to stay connected on LinkedIn.

Generative AI will impact all industries. 
 
This is real-time voice cloning, lip syncing, language translation, face tracking, etc. 
 
Amazing. 

Source: Flawless 
____ 
#ai #machinelearning #datascience

Miguel Fierro

OpenAI, the company behind ChatGPT, is valued at $29 Billion.  
 
Multiple firms are interested in investing. 
 
It makes it the 6th top-valued unicorn: SpaceX ($137B), Stripe ($74B), Epic Games ($32B), Databricks ($31B), Fanatics ($31B), and OpenAI ($29B) 
 
OpenAI was founded in 2013 by Elon Musk and venture capitalist Sam Altman as an artificial intelligence research lab. 
 
OpenAI is in my opinion one of the best executors of transforming AI research into a product.  
 
This is actually my career journey, I specialized in transforming research into products, and it is one of the most difficult things you can do (if not the most) 
 
There are very few companies that are able to achieve this. OpenAI is one, and Microsoft is another one. Maybe that's why their partnership is working so well. 
 
It's super exciting to be working in AI these days.  

If you love machine learning and want to switch to a Data Science career, join https://lnkd.in/dfNXNDa4 to receive in your inbox every Sunday tricks and strategies to help you.
____ 
#ai #machinelearning #datascience

No alternative text description for this image

10 reasons why having a GitHub portfolio can be beneficial for your career in Data Science:

1. It shows that you are actively engaged in learning and improving your skills.
2. It demonstrates your ability to work with version control and collaborate on projects.
3. It allows potential employers to see your work and code in action.
4. It can help you network with other professionals in the field.
5. It can help you get feedback on your projects from other developers.
6. It can serve as a reference for your past work and accomplishments.
7. It can help you stand out in a competitive job market.
8. It can help you track your own progress and development as a data scientist.
9. It can serve as a portfolio when applying for academic programs or grants.
10. It can be a useful resource when preparing for technical interviews.

If you like this content, consider joining my weekly email newsletter where I share ideas and strategies on how to switch your career to Data Science https://lnkd.in/dfNXNDa4
____
#ai #datascience #machinelearning

People say that ChatGPT is not a threat to Google. 
Because it is not 100% accurate. 
They don't understand the concept of the innovator's dilemma. 
 
The Innovator’s Dilemma is a book published in 1997 by Clayton Christensen. 
 
It explains why established companies fail to respond to disruptive products and eventually die or are left behind. 
 
A disruptive product with lower performance can win because it addresses the needs of a different market segment that established companies are not currently serving.  
 
These customers are willing to trade off performance for other attributes, such as lower cost, speed, or simplicity. 
 
Disruptive innovations gradually improve and eventually disrupt the established market, overtaking it. 

The dilemma comes because established companies may not see a large enough market for them, and thus will miss the potential of this disruptive technology. 
  
A recent example of a disruptive technology that is eating an established market is electric cars.  
 
Initially, they had a shorter range and a lack of charging infrastructure. Electric cars appealed to a market segment of environmentally conscious consumers and urban dwellers who do not need to drive long distances. When the electric car improves, it can replace gasoline cars. 
 
This is exactly what is happening with ChatGPT.  
 
The information that ChatGPT provide is not accurate 100% of the time, but there is a market segment that is willing to trade accuracy for speed.  
 
What's the solution to the innovator's dilemma?  
 
Acquiring the company or funding subsidiaries. 

If you liked this story, consider joining https://lnkd.in/dfNXNDa4
____ 
#innovation #future #ai

The Transformer architecture has impacted NLP massively and is also used in other areas. 

In this post, I'll explain how Transformers are applied to sequential recommendation systems and share a code example.

In sequential recommendation systems, the objective is to predict the next item the user will interact with. 

Traditionally, this problem has been solved with LSTMs or CNNs.

LSTMs are good at learning relationships in long sequences but are slow because of their recurrent nature. 

CNNs, are not as good at learning relationships in long sequences but can work in parallel. 

Transformers get the best of both worlds, they are able to learn long-term relationships and work in parallel. 

The key is the multi-head self-attention module. 

The attention mechanism learns what items are important in the sequence of actions. 

In this notebook, we show how to use two attention-based sequential recommenders, SASRec and SSEPT: https://lnkd.in/dWmwCUDX 

Hope this is useful.
____
#ai #machinelearning #recommendationsystems

Open article: GitHub - microsoft/recommenders: Best Practices on Recommendation Systems by github.com

GitHub - microsoft/recommenders: Best Practices on Recommendation Systems

As of today, we can confirm that the rumors you've been hearing are true. ChatGPT is coming to Azure.

Today we announce the General Availability of the Azure OpenAI Service. 

This includes GPT-3.5, Codex, DALL-E 2, and soon ChatGPT.

See the official announcement: https://lnkd.in/dj7JWMam
____
#ai #machinelearning #azure

This week has been hard for people in tech. 
 
10k layoffs at Microsoft, 18k at Amazon, and 12k at Google. 
 
Here are some lessons I learned from all this. 
 
Anyone can be laid off even if they: 
- Work late hours. 
- Go above and beyond. 
- Deliver high-quality projects. 
- Do everything they are told to do. 
 
The approach I take is different: 
- I work late hours for my personal development. 
- I go above and beyond for the things I learn. 
- I work to get better every day. 
- I find purpose in my life. 
 
By spending time becoming the best AI professional I could be: 
1. I serve my company and its clients better. 
2. If I'm laid off, I have a better chance of getting a new job. 
 
The key idea is this: 
 
Don't let your job control your life. 
 
Make your job a platform for your personal growth. 
 
Tomorrow I'll share with my email community how to navigate these difficult times. Get the content here: https://lnkd.in/dfNXNDa4. 
____ 
#layoffs #personaldevelopment #careeradvice

There is a new extension that uses ChatGPT to generate summaries of YouTube videos.

I've been playing with it and it works well when the video is not very long.

Pretty useful!

See the info in the first comment.
____
#ai #technology #future

A new app helps detect whether a text has been generated by ChatGPT or by a human. It's called GPTZero. Let me explain how it works. 
 
The app measures a metric called perplexity in the text. 
 
Formally, for the task of predicting the next word in a document, perplexity is defined as the average per-word log-probability (lower is better). 
 
Human-generated texts tend to have a higher average perplexity (average perplexity across sentences), and also higher burstiness (the perplexity is not uniform across sentences).  
 
Even though it is not perfect, it gives people a measurement that can be used to analyze whether the text has been generated with AI or not. 
 
The image below has an average perplexity of 43.1 and not quite a uniform perplexity distribution. This was from a post that I slightly modified from a ChatGPT output. 
 
See more information below. 
____ 
#ai #machinelearning #nlp

Failure is the backprop signal that helps us become better.

Everybody fails.

Even robots 🤖🤖🤖

Happy new year folks!

Vamos!🦾🦾🦾
____
#jokeoftheday #robotics 

The method used to retrain the GPT model is a large part of ChatGPT's success. 
  
It is called Reinforcement Learning from Human Feedback (RLHF). 
 
Here is a short summary. 
 
All language models are trained in two phases: pre-training and finetuning.  
 
Pretraining is training a network on unlabeled data. For example, we can use Wikipedia and pretrain the network to predict the next word based on the previous word.  
 
This is a cheap process because there is no human labeling involved. It makes the network learn the structure of the English language. 
 
Finetuning is retraining the network again on labeled data. The network already understands English (it has pretrained weights) and we can use traditional supervised learning for tasks like text classification, machine translation, or text generation. 
 
This process is expensive because it involves humans labeling the dataset. It makes the network learn a specialized task, such as providing an answer to a question. 
 
RLFH is a finetuning method with the objective of following the user’s instructions helpfully and safely. 
 
In RLHF, humans rate multiple questions and answer pairs, and rank them on quality. 
 
This generates a reward function that is used to finetune the network to maximize the cumulative future reward. In other words, it produces answers that are helpful, safe, and of high quality. 
 
I hope this is useful. 

If you love machine learning and want to switch to a Data Science career, join https://lnkd.in/dfNXNDa4 to receive in your inbox every Sunday tricks and strategies to help you.
____ 
#ai #machinelearning #datascience

For every layoff post we see on LinkedIn,  
there are likely 10 more people going through the same experience,  
but haven't shared anything. 
 
They also: 
- May be struggling financially. 
- May feel lost on what to do next. 
- May be forced to leave their country. 
- May feel helpless, anxious, or depressed. 
 
They may be silent but they need as much help and support as everybody else. 
____ 
#layoffs #career #job #opentowork

Heartbroken by the layoffs at Microsoft and Amazon. Let me help.  
 
I have 100s of recruiters in my network, this message can reach them. 
 
Here's what to do: 
 
Reply to this message with: 
 
1. Profession and years of experience in the company. 
2. Where you are based. 
3. The best project you did and the results you achieved. 
4. How to reach you. 
 
Please share to help the tech community. 
____ 
#layoffs #microsoft #amazon #opentowork #amazonlayoffs #microsoftlayoffs

If you are interested in object detection, here is a short summary of the state-of-the-art algorithm YOLOv7.

YOLOv7 is a very interesting object detection solution with really good inference speed and accuracy.

YOLOv7 comes with 4 main improvements:
- Extended Efficient Layer Aggregation: layer structure that produces easier-to-optimize gradients and learns more diverse features.
- Model scaling for concatenation-based models: object detection internal modules have scale-up and scale-down components to control the depth and width of the network. In YOLOv7 the authors scale the network depth and width in concert while concatenating layers together.
- Re-parameterization Planning: Averaging the model weights to create a more robust model.
- Auxiliary Head Coarse-to-Fine: In the standard YOLO the head is the final block of the network used to make the prediction. In YOLOv7, auxiliary heads are added in the middle of the network to improve the prediction.

Check out the papers and code in the comments below.
____
#ai #machinelearning #computervision

When you take git to a new level. 
 
___ 
#jokeoftheday #programmingmemes

I was talking to an xMicrosoft friend about roles in companies creating AI products. We both agreed that Software Engineers are super valuable in machine learning teams.  
 
Here are some reasons to consider SE for AI teams: 
 
- They typically know how to create APIs and user interfaces. 
- They typically master multiple programming languages. 
- They can bring good software practices to the team. 
- They can help Data Scientists with infrastructure. 
- They can optimize algorithms. 
 
People that have experience in Software Engineering and then switched their careers to Data Science are going to be in high demand. 
____ 
#ai #machinelearning #softwareengineering

If you're attempting to change careers to Data Science, there is a simple solution to show your value: setting up a GitHub portfolio. On this website, you can find the solution to +500 Kaggle competitions. 
 
If there is one thing that has improved my coding ability and problem-solving ability is looking at other successful projects. 
 
Here is the website: https://lnkd.in/dkxeEQY6 
 
I hope this is useful. 
 
Want to switch to a Data Science career? Subscribe to https://lnkd.in/dfNXNDa4 for weekly tips. 
____ 
#ai #machinelearning #datascience  

Many people think that Juypter notebooks are not for production ML. 
 
But they can be useful assets if you take advantage of them. 
 
Here is how: 
 
- Facilitate collaboration. 
- Help with experimentation. 
- Good vehicle to communicate results. 
- Facilitate data exploration and visualization. 
 
Misuse of notebooks includes: 
 
- Using notebooks as the primary method for deploying ML models. 
- Having supporting code in the notebook instead of in a library. 
- Not testing the code inside the notebook. 
- Having sensitive data in the notebook. 
 
Jupyter notebooks can be the front-end of your machine learning solution. 
 
Libraries can be the back-end. 
____ 
#ai #machinelearning #datascience

The next two years are going to be difficult for people in tech. 
 
Layoffs, instability, fewer job opportunities... 
 
Building relationships can be your safety net. 
 
They say your network is your net worth, here are some advantages of building relationships: 
 
- More job security. 
- Advance in your career. 
- New career opportunities. 
- Develop new skills and knowledge. 
- Provide support and a sense of community. 
 
Tomorrow I'll share with my email community some strategies I've learned to build relationships that can help you in your career. Get the content here: https://lnkd.in/dfNXNDa4 
____ 
#innovation #future #personaldevelopment #careeradvice

The current layoffs can be a wake up call for many to switch their careers to a job they like.  
Now it's more important than ever to find the time and make the time. 
Some ideas to make sure you manage time: 
 
1. Get creative with your time:  
Combine your personal development with other activities. For example, you can listen to an AI podcast while you work out. 
 
2. Make use of downtime:  
Take advantage of small pockets of free time throughout the day to learn something new. For example, use your commute time to listen to an educational podcast. 
 
3. Block your calendar: 
Treat your personal development time as any other important appointment. For example, schedule non-negotiable 1-1 with yourself.  
 
4. Be accountable: 
Make yourself accountable or find someone who is also committed to personal development to help you. For example, a partner or friend can hold each other accountable. 
 
5. Prioritize:  
Evaluate your schedule and prioritize your personal development time. For example, this may mean saying no to other commitments or delegating tasks to free up time. 
 
In summary: 
- Get creative with your time. 
- Make use of downtime. 
- Block your calendar. 
- Be accountable. 
- Prioritize. 
 
Vamos!!! 🦾🦾🦾 

Looking to transition into a Data Science career? Subscribe to https://lnkd.in/dfNXNDa4 for weekly updates.
____ 
#layoffs #careerdevelopment #careerswitch #job

Taplio

Miguel Fierro's Linkedin Analytics

Miguel Fierro

What is Miguel talking about?

Who is engaging with Miguel

Miguel Fierro's Best Posts (last 30 days)

Want to drive more opportunities from LinkedIn?

Taplio

.css-1jgreyo{background-image:linear-gradient(284deg, rgba(39,43,72,1) 0%, rgba(33,47,183,1) 100%);color:transparent;-webkit-background-clip:text;background-clip:text;}Miguel Fierro's Linkedin Analytics

Miguel Fierro

What is Miguel talking about?

Who is engaging with Miguel

Miguel Fierro's Best Posts .css-qqfgvy{font-size:var(--chakra-fontSizes-md);}(last 30 days)

Want to drive more opportunities from LinkedIn?

Miguel Fierro's Linkedin Analytics

Miguel Fierro's Best Posts (last 30 days)