Jack Dąbrowski

Jack Dąbrowski

Is currently leading the AI Research department at Synerise. He is responsible for development of unsupervised behavioral deep learning models on multimodal, heterogeneous web-scale data.

📍 Warsaw, Poland

BaseModel vs HSTU for sequential recommendations

In May 2024, a preprint titled "Actions Speak Louder than Words: Trillion-Parameter Sequential Transducers for Generative Recommendations" [1] was posted by Meta AI researchers to ArXiv. The preprint introduced a novel recommender, referred to as 'HSTU' which stands for “Hierarchical Sequential Transduction Units” promising new state-of-the-art

BaseModel vs TIGER for sequential recommendations

In December 2023, a paper titled "Recommender Systems with Generative Retrieval" [1] was presented by researchers at NeurIPS 2023. This paper introduced a novel recommender system, referred to as 'TIGER,' which garnered significant attention due to its innovative approach and state-of-the-art results in sequential recommendation tasks.

Temporal data splits

Understanding human behavior is a complex and challenging task. Behavioral modeling domain spanning tasks such as recommendation, propensity-to-buy, or LTV, require solutions tailored to domain's specific requirements. However, solutions are often copy-pasted from other domains. In behavioral modeling, we aim to predict future behaviors based on user'

EMDE vs Multiresolution Hash Encoding

When we created our EMDE algorithm we primarily had in mind the domain of behavioral profiling. For this reason, we were looking for a method to aggregate multiple events represented by various types of embeddings. However recently, a similar method was developed in the visual domain for two-dimensional data. Researchers

Project BaseModel

In this article we introduce BaseModel, a tool which makes the most of event data reflecting user behavior. Working with clients in many different industries, we have observed that large company data is often underutilized. Here we will discuss where our research on BaseModel arises from, how we meet strict

Fourier Feature Encoding

Pre-processing raw input data is a very important part of any machine learning pipeline, often crucial for end model performance. What is more, different fields of ML require different methods to represent relevant information from the input domain in machine-readable format – as numerical vectors. In image processing, we can use

Efficient integer pair hashing

In data science, we sometimes need to calculate hash functions of unsigned 32-bit integer tuples. This need can arise in simple use cases such as efficient pairs counting, but also in approximate data structures which require fast hashing. Such data structures provide a tradeoff between accuracy and speed/memory storage:

How we challenge the Transformer

Having achieved remarkable successes in natural language and image processing, Transformers have finally found their way into the area of recommendation. Recently, researchers from NVIDIA and Facebook AI joined forces to introduce Transformer-based recommendation models described in their RecSys2021 publication Transformers4Rec: Bridging the Gap between NLP and Sequential / Session-Based Recommendation,

EMDE Illustrated

EMDE – motivations, explanations and intuitions In this article we provide some intuitive explanations of our objectives and theoretical background of the Efficient Manifold Density Estimator (EMDE). We describe our inspirations and thought process which led to the formulation of the algorithm. Then, we dive further into technicalities to show how

Towards a multi-purpose behavioral model

In various subfields of AI research, there is a tendency to create models which can serve many different tasks with minimal fine-tuning effort. Stanford researchers introduced the concept of foundation models, which are trained on massive datasets and can be adapted to different downstream applications. We may think this way