“Yoel Zeldes worked in my Engineering organization at Fortscale as part of the 'engine' group developing machine learning / statistical, behavioral algorithms to combat cyber security. During his time at Fortscale, Yoel proved to be a sharp and analytical guy that can face any thinking challenge and come with the best solution. Yoel came to Fortscale to focus on research and algorithms and quickly became an expert in the Fortscale algorithm and developed several great enhancements and improvements for the algorithm. Outside the algorithm development, Yoel did not hesitate to take ownership of some challenging parts of the Fortscale product and has developed features and capabilities in multiple areas with high professionalism and quality. On the personal side, Yoel has a great spirit at work and life. He is a great team member and at Fortscale he contributed to the overall atmosphere and the fun feeling of coming to work”
About
Over 13 years of experience as a software engineer and algorithm developer in various…
Contributions
-
What are common NLP tasks and applications?
There are two types of summarizing systems: extractive and abstractive. Extractive summaries can only contain words that appear in the original text, while abstractive summaries can contain new words as well. With this flexibility, an abstractive summarizing system can generate better summaries, at the risk of introducing severe errors.
-
How can you create an algorithm that is fair for all users?
When working with large language models, fairness and bias often require the models to reject some user instructions. Model trainers must ensure that the model has a satisfactory true positive rate, that is, that it refuses to follow harmful instructions. However, false positives are inevitable, meaning that the model will refuse to follow some legitimate instructions. This is a risk the model designers have to take into account.
-
How can you create an algorithm that is fair for all users?
It is indeed possible to use algorithmic fairness to your advantage, or at least to distinguish you from your competitors. Anthropic is a good example: They are committed to creating helpful, honest, and harmless large language models (https://www.anthropic.com/index/claudes-constitution). A fair and unbiased model can help shape this narrative, which is important for many businesses that want to consume LLM outputs.
Activity
-
Today we’re releasing Jamba 1.5 Mini and Jamba 1.5 Large. At AI21 Labs we were never ones to follow convention. We founded the company based on the…
Today we’re releasing Jamba 1.5 Mini and Jamba 1.5 Large. At AI21 Labs we were never ones to follow convention. We founded the company based on the…
Liked by Yoel Zeldes
-
We’re proud to share that we made The Cloud Awards 2024 #AIAwardsShortlist in not one, but FOUR categories! 🥇AI Startup of the Year 🥇AI Innovation…
We’re proud to share that we made The Cloud Awards 2024 #AIAwardsShortlist in not one, but FOUR categories! 🥇AI Startup of the Year 🥇AI Innovation…
Liked by Yoel Zeldes
Experience
Education
Publications
-
Technical Report: Auxiliary Tuning and its Application to Conditional Text Generation
We introduce a simple and efficient method, called Auxiliary Tuning, for adapting a pre-trained Language Model to a novel task; we demonstrate this approach on the task of conditional text generation. Our approach supplements the original pre-trained model with an auxiliary model that shifts the output distribution according to the target task. The auxiliary model is trained by adding its logits to the pre-trained model logits and maximizing the likelihood of the target task output. Our method…
We introduce a simple and efficient method, called Auxiliary Tuning, for adapting a pre-trained Language Model to a novel task; we demonstrate this approach on the task of conditional text generation. Our approach supplements the original pre-trained model with an auxiliary model that shifts the output distribution according to the target task. The auxiliary model is trained by adding its logits to the pre-trained model logits and maximizing the likelihood of the target task output. Our method imposes no constraints on the auxiliary architecture. In particular, the auxiliary model can ingest additional input relevant to the target task, independently from the pre-trained model's input. Furthermore, mixing the models at the logits level provides a natural probabilistic interpretation of the method. Our method achieved similar results to training from scratch for several different tasks, while using significantly fewer resources for training; we share a specific example of text generation conditioned on keywords.
Other authorsSee publication -
Deep density networks and uncertainty in recommender systems
KDD (Knowledge Discovery and Data Mining) - best paper award (AdKDD & Target Ad workshop)
Building robust online content recommendation systems requires learning complex interactions between user preferences and content features. The field has evolved rapidly in recent years from traditional multi-arm bandit and collaborative filtering techniques, with new methods employing Deep Learning models to capture non-linearities. Despite progress, the dynamic nature of online recommendations still poses great challenges, such as finding the delicate balance between exploration and…
Building robust online content recommendation systems requires learning complex interactions between user preferences and content features. The field has evolved rapidly in recent years from traditional multi-arm bandit and collaborative filtering techniques, with new methods employing Deep Learning models to capture non-linearities. Despite progress, the dynamic nature of online recommendations still poses great challenges, such as finding the delicate balance between exploration and exploitation. In this paper we show how uncertainty estimations can be incorporated by employing them in an optimistic exploitation/exploration strategy for more efficient exploration of new recommendations. We provide a novel hybrid deep neural network model, Deep Density Networks (DDN), which integrates content-based deep learning models with a collaborative scheme that is able to robustly model and estimate uncertainty. Finally, we present online and offline results after incorporating DNN into a real world content recommendation system that serves billions of recommendations per day, and show the benefit of using DDN in practice.
Other authorsSee publication -
On-line fair allocations based on bottlenecks and global priorities
ICPE (International Conference on Performance Engineering)
System bottlenecks, namely those resources which are subjected to high contention, constrain system performance. Hence effective resource management should be done by focusing on the bottleneck resources and allocating them to the most deserving clients. It has been shown that for any combination of entitlements and requests a fair allocation of bottleneck resources can be found, using an off-line algorithm that is given full information in advance regarding the needs of each client. We extend…
System bottlenecks, namely those resources which are subjected to high contention, constrain system performance. Hence effective resource management should be done by focusing on the bottleneck resources and allocating them to the most deserving clients. It has been shown that for any combination of entitlements and requests a fair allocation of bottleneck resources can be found, using an off-line algorithm that is given full information in advance regarding the needs of each client. We extend this result to the on-line case with no prior information. To this end we introduce a simple greedy algorithm. In essence, when a scheduling decision needs to be made, this algorithm selects the client that has the largest minimal gap between its entitlement and its current allocation among all the bottleneck resources. Importantly, this algorithm takes a global view of the system, and assigns each client a single priority based on his usage of all the resources; this single priority is then used to make coordinated scheduling decisions on all the resources. Extensive simulations show that this algorithm achieves fair allocations according to the desired entitlements for a wide range of conditions, without using any prior information regarding resource requirements. It also follows shifting usage patterns, including situations where the bottlenecks change with time.
Other authors -
Projects
-
AnotherDatum
- Present
A blog about data science and software engineering that I'm writing. Code can be found here: https://github.com/yoel-zeldes/yoel-zeldes.github.io
-
www.liatzeldes.com
-
I built this website for my sister.
The code can be found here:
https://github.com/yoel-zeldes/liat-website
Honors & Awards
-
Magna Cum Laude
Hebrew University Of Jerusalem
In addition, included in Dean's List Award twice (in second and third years).
Languages
-
Hebrew
-
-
English
-
Recommendations received
2 people have recommended Yoel
Join now to viewMore activity by Yoel
-
את האפליקציה הזאת כתבתי בחאן יונס, וזה בכלל התחיל כבדיחה. לכל אחד בצוות יש פק"ל מבצעי, קלע, נגביסט, חובש וכו', ופק"ל חברתי. הבשלן, זה שמדרבק קפה כל…
את האפליקציה הזאת כתבתי בחאן יונס, וזה בכלל התחיל כבדיחה. לכל אחד בצוות יש פק"ל מבצעי, קלע, נגביסט, חובש וכו', ופק"ל חברתי. הבשלן, זה שמדרבק קפה כל…
Liked by Yoel Zeldes
-
⌚ Time Management Hack: Embrace Your Productivity Peaks (and Valleys) Ever feel like you're just banging your head against a wall trying to get work…
⌚ Time Management Hack: Embrace Your Productivity Peaks (and Valleys) Ever feel like you're just banging your head against a wall trying to get work…
Shared by Yoel Zeldes
-
The future belongs to 𝐞𝐟𝐟𝐢𝐜𝐢𝐞𝐧𝐭 𝐥𝐨𝐧𝐠-𝐜𝐨𝐧𝐭𝐞𝐱𝐭 LLMs. Join me for a walk. -- We all know (and even more so if you're following…
The future belongs to 𝐞𝐟𝐟𝐢𝐜𝐢𝐞𝐧𝐭 𝐥𝐨𝐧𝐠-𝐜𝐨𝐧𝐭𝐞𝐱𝐭 LLMs. Join me for a walk. -- We all know (and even more so if you're following…
Liked by Yoel Zeldes
-
🚀 Want to become an outstanding employee? First, you need to learn the difference between an outstanding employee and a technically strong but…
🚀 Want to become an outstanding employee? First, you need to learn the difference between an outstanding employee and a technically strong but…
Shared by Yoel Zeldes
-
Have you used a search engine other than Google in the past 20 years? Is it possible that there's been no successful innovation in search outside of…
Have you used a search engine other than Google in the past 20 years? Is it possible that there's been no successful innovation in search outside of…
Liked by Yoel Zeldes
-
There's a new boy in "buzzword" town called "Overemployment". It essentially means individuals can handle more than one full-time job at the same…
There's a new boy in "buzzword" town called "Overemployment". It essentially means individuals can handle more than one full-time job at the same…
Liked by Yoel Zeldes
-
Hack by popular demand! Open-Source Agent #Hackathon with Andrew Ng & DeepLearning.AI at the AGI House this Saturday, August 24th! Join us for…
Hack by popular demand! Open-Source Agent #Hackathon with Andrew Ng & DeepLearning.AI at the AGI House this Saturday, August 24th! Join us for…
Liked by Yoel Zeldes
Other similar profiles
Explore collaborative articles
We’re unlocking community knowledge in a new way. Experts add insights directly into each article, started with the help of AI.
Explore More