recall morphemes are the smallest meaningful units of a word.

morphological parsing is the act of getting morphemes: cats => =cat s=o

1. stem +
2. affix

stemming

stemming just chops off the morpheme affixes; leaving the stems. “heights” => “heigh”. without lemmatization.

This increases recall (more stuff is caught we want to catch) at he cost of precision (what we catch is probably lots of false positives).

Languages with complex cojugation or morphology, this can’t work because you can’t just chop.

• Solution for Satelite Assignment Problem
• MARL for Combinatorial Optimization

Take X-Rays and generate clinical reports.

Method

encoder decoder architectures

Encoder

ConViT: convolutional vision transformer. Special thing: we swap out the attention

Double Weighted Multi-Head Attention

We want to force the model to focus on one thing, so we modulate the model based on the weights of other: if one head is big, we make the other head small.

where \(w_{\cos i} = \frac{\sum_{i}^{} \cos \qty (att_{a}, att_{base})}{N}\)

\begin{equation} w = w_{a} \cdot (1- w_{\cos i}) \end{equation}

simple games

constituents

• agent \(i \in X\) the set of agents.
• joint action space: \(A = A’ \times A^{2} \times … \times A^{k}\)
• joint action would be one per agent \(\vec{a} = (a_{1}, …, a_{k})\)
• joint reward function \(R(a) = R’(\vec{a}), …, R(\vec{a})\)

additional information

prisoner’s dilemma

traveler’s dilemma

• two people write down the price of their luggage, between 2-100
• the lower amount gets that value plus 2
• the higher amount gets the lower amount minus 2

joint policy agent utility

for agent number \(i\)

Key idea: multi-modality, when leveraged well, leads to faster convergence.

Data Availability

Health and health sensing requires labels, but health signals require specialist knowledge + broader context to label.

• typical image labeling: 0.05/label
• medical imaging: 4.00/label

Even if want to automate the study, we need to Kyntic style strap a thing to a person and have soft labels that we align with raw sensor data..

Instead, Do Time-series

Instead: run proxy self-supervised studies into the future—pretraining on a shit tone of sensor data just as timeseries regressing into the future without any labels.