Transfer & Self-Supervised Learning

Why this map looks & behaves the way it does

1. Compare, don't just list. Studying two cases side by side and naming the dividing question teaches the distinction far better than reading each alone (Gentner's structure-mapping; Bransford & Schwartz 1999 "contrasting cases"; Rittle-Johnson & Star 2007). → Every ⚖ node is a head-to-head: hover for the two columns and the question that splits them.
2. Commands beat descriptions. Phrasing material as something to do ("freeze the early layers", "mine hard negatives") engages active processing more than passive statements (generation effect, Slamecka & Graf 1978; enactment effect). → Cues are written as instructions, not facts.
3. Mind maps measurably help. Better week-later recall (Farrand et al. 2002); meta-analyses over 55+ studies find consistent gains, g ≈ 0.58 (Nesbit & Adesope 2006; Schroeder et al. 2018).
4. Retrieval practice beats re-studying (Karpicke & Blunt 2011, Science). → Recall mode blurs the details; say them aloud, click to check.
5. Color carries meaning (Ozcelik et al. 2009; Dzulkifli & Mustafar 2013). → Each branch colour marks a stage; gold marks a comparison.
6. Chunk & hierarchy, big-picture first (Cowan 2001; Bower et al. 1969; Mayer's segmenting). → 5 branches, opens collapsed, detail on demand. And space it out (Cepeda et al. 2006): re-run recall mode today, tomorrow, in 3 days.