Other

Motivation • Drug legal status determined not only chemically, but also historically. • Can we predict it? • Can we predict abuse potential of substances? This work • Docking Bad dataset ◦ Substances with WHOCC ATC classification N (= nervous system) (WHO Collaborating Centre for Drug Statistics Methodology, 2024) ◦ Substances with the DEA Controlled Substances status (Drug Enforcement Administration, n.d.) ◦ Features collected using PubChem (Kim et al., 2023) + ChEMBL (Mendez et al., 2019) ◦ SMILES, receptor activity, … • ATC classification prediction • DEA schedule prediction

Motivation • “Music is a language” • Jazz especially ◦ rhythmic, harmonic, and melodic grammar and vocabulary • So, let us try NLP • We will focus on harmony ◦ Symbolic, structured ◦ Musicians think about it consciously (more frequently than, e.g., rhythm) This work • Key prediction from chord BoWs • Root motion n-gram Markov models • Generation!

Motivation • Graphs used as an abstract language throughout modern STEM • Can encode discrete structures — molecules, social networks, etc. Problem setup A graph \( G \) over \( n \) vertices is represented as: \( G = (X,A), X \in \mathbb{R}^{n \times d}, A \in \{0, 1\}^{n \times n} \) Goal: learn a distribution over graphs and generate novel, faithful samples. Classical generative models and their limitations: • VAEs (Simonovsky and Komodakis, 2018) — expensive graph-matching for permutation invariance • GANs (Wang et al., 2018) — mode collapse (reduction of diversity) • Normalising flows (Luo, Yan, and Ji, 2021) — constrained to bijective mappings Diffusion models have emerged as a compelling alternative.

My transcriptions of E.S.’ solos that I think are crucial for understanding the artist. Some are complete (Decade – Leucocyte, I Mean You Introduction – Jazz Baltica ‘99), others (Say Hello To Mr. D., Somewhere Else Before, …) only focusing on certain things I was looking for at the time.

In Euclidean data clustering, we seek to assign input data points into clusters, such that the maximum distance between points within the same cluster is minimized (or alternatively, such that the maximum distance between points from their cluster’s center is minimized). Through various reductions of graph problems (\(k\)-coloring, vertex cover), it is possible to show that not only is this task in \(\ge 3\) dimensions NP-complete, but also its approximations are NP-complete. Our aim in this project will be to improve known approximation factors for NP-completeness or polynomial complexity.

Consider a finite set of vertices \(V\) and suppose that we only know distances between some of them. In other words, consider a random graph \( G \) over \(V\) and a function \(d: E(G) \to \mathbb{R}^+ \cup {0}\). What properties of \(G\) make it possible to uniquely reconstruct the configuration of \(V\) (up to isometry)? We will show that the minimum degree \(\ge 2\) is alone sufficient with high probability.

We have recovered a CD from the abandoned Milada squat house.

A (commutative, unital) ring is the 3-tuple \( (R, +, \cdot) \) such that \( (R, +) \) is an Abelian group, \( (R, \cdot) \) is a commutative monoid, and the two satisfy mutually distributive laws. An \( R \)-module is an Abelian group tied to a scalar multiplication ring \( R \), satisfying mutual distributivite laws and multiplicative compatibility.

A (commutative, unital) ring is the 3-tuple \( (R, +, \cdot) \) such that \( (R, +) \) is an Abelian group, \( (R, \cdot) \) is a commutative monoid, and the two satisfy mutually distributive laws. \( S \subset R \) is a subring of \( R \) if there exists a ring morphism \(R \to S \). An ideal \(I \subset R \) is a additive subgroup of \( R \) that absorbs all multiplication: \( \forall i \in I, r \in R: ir \in R \).