SU-CS329H OCT072026
Last edited: October 10, 2026Key Sequence
Notation
New Concepts
Important Results / Claims
The same train test val lecture, but this time now.
| Hold out | to examine | keeping in mind that |
|---|---|---|
| comparisons | new pairs in known entities | duplicates and reverse pairs are together |
| users | preferences in new users | find available features or history |
| prompts / context | preferences in new content | group related responses by prompt |
Utility Representations
Last edited: October 10, 2026What are some ways we can get \(r_{i}\) as in the features that goes into some kind of way of getting probabilities for preferences?
Brady-Terry
\begin{equation} p\qty(i \succ j) = \sigma \qty(z), z_{ij} = s_{i} - s_{j} \end{equation}
factor model
- itemwise: \(z_{ij} = u_{i} v_{j} + \beta_{j}\)
- pairwise: \(z_{i,a,b} = u_{i}\qty(v_{a} - v_{b}) + \qty(\beta_{a}- \beta_{b})\)
see factor model
Assume a bilinear structure of preferences from above
- \(u_{i} = G_{\psi}\) <- some thing about users
- \(v_{x,a} = E_{\phi}\qty(x,a)\) <- BERT or similar
- \(r\qty(i,x,a) = u_{i}^{T} v_{x,a} + b_{\eta}\qty(x,a)\)
where \(u_{i}\) is the user vector, \(x\) the context, \(a\) the document ID, and \(b_{\eta}\) is the baseline document preferences.
Agents Index
Last edited: October 10, 2026Lectures
cool thing
https://trafilatura.readthedocs.io/en/latest/
project thoughts
mixture of depth + speculative decoding
elimination by aspects
Last edited: September 9, 2026Each item is a set of “aspect.” Aspect \(\alpha\) has weight \(w\qty(\alpha) > 0\). And the way to get probability distribution of which items to pick:
given menu \(S\), discard aspects shared by every item in \(S\). then:
- choose an aspects with probability \(\propto w\qty(\alpha)\)
- keep only the items that remain, repeat until the list is broken down
Insight: “similar thing being added increasingly should not as much more decision share. Red bus + green ice cream adds more decision share than Red bus + blue bus + green ice cream which adds more than Red + Blue + green …. + purple bus + green ice cream”
Fechner and Quadrupel Condition
Last edited: September 9, 2026Fechner / strong utility: there are \(u : X \to \mathbb{R}\) and \(F: \mathbb{R} \to \qty(0,1)\), continuous, strictly increasing, and \(F\qty(-t) = 1 - F\qty(t)\), and
\begin{equation} P\qty(j \succ j’) = F\qty(u_{j} - u_{j’}) \end{equation}
Quadruple Condition: for all choices of \(j, j’, j’’, j’’’\), we have:
\begin{equation} P\qty(j \succ j’) \geq P\qty(j’’ \succ j’’’) \Leftrightarrow P\qty(j \succ j’’) \geq P\qty(j’ \succ j’’’) \end{equation}
Solvability: for \(j, j’, j’’\) and \(\lambda \in [0,1]\) with \(P\qty(j \succ j’) \leq \lambda \leq P\qty(j \succ j’’)\), some \(j’’’\) has \(P\qty(j \succ j’’’) = \lambda\)
