Trace a learning and query path
Follow one feature value into DNA memory and back into a query score.
Course lessons · 2 of 5
Evaluate a coefficient update and explain why a compact feature sketch is not automatically lossless memory.
Useful background: Read a build as a system
Encoding, learning, and querying are different operations
The SEFERIM DNA memory record maps an item through a normalized hash and sinusoidal features. Learning changes a retained coefficient vector. Querying combines query features with those coefficients in a dot-product-like score. These steps have different inputs and purposes.
The historical record contains an additive update and a later blending update. The blend retains a fraction of the old coefficient and incorporates a scaled new signal. Changing between the two rules changes accumulation and forgetting behavior.
Follow where information is preserved or combined
A finite hash can map different items to the same value. Combining multiple encoded items into a short coefficient vector also merges their contributions. Neither a hash nor a feature similarity score alone provides lossless reconstruction of every original item.
A surrounding system can retain source files, identifiers, event logs, or substrate addresses separately. Good architecture names those stores explicitly. To measure memory usefulness, test retrieval accuracy, interference, capacity, and forgetting on a defined dataset.
Update one coefficient in a teaching sketch
1Set the retained value and input
Let the old coefficient be 0.2 and the already-computed learning signal s be one. Choose α = 0.25.
2Weight old and new contributions
The retained part is 0.75 × 0.2 = 0.15. The new contribution is 0.25 × 1 = 0.25.
3Add the two contributions
The new coefficient is 0.4. This one-coordinate example illustrates the blend without claiming to reproduce the entire memory pipeline.
Make the idea respond
The normalized aggregate can reproduce the one-coordinate blend: set value A to 0.2, value B to 1, weight A to 3, weight B to 1, and weight C to 0. The normalized weights become 0.75 and 0.25, and the score is 0.4. This setup illustrates one blend; it does not implement item hashing, retained DNA memory, or the unrestricted dot product of 11 in the question.
Change one input at a time. Predict what should happen, then compare the computed result with your prediction.
Teaching model · computed in your browser
Build a weighted score
Change component values and weights. Watch normalization turn them into a single, explainable score.
vᵢ is a component value from 0 to 1; wᵢ is a nonnegative raw weight. Divide every weight by their positive total before combining the values. These illustrative values are not measurements from a live AI system.
Try this: double all three weights. The normalized weights and total score should stay the same. Then set all weights to zero.
The default worked example is shown below.
Normalize before adding: divide each raw weight by 6. Then S = 0.4 + 0.2 + 0.0667 = 0.6667. Multiplying every raw weight by the same positive factor leaves the result unchanged. Choosing weights expresses a modeling decision; it does not validate what a score means.
| Component | Value | Raw weight | Normalized weight | Contribution |
|---|---|---|---|---|
| Component A | 0.8 | 3 | 0.5 | 0.4 |
| Component B | 0.6 | 2 | 0.3333 | 0.2 |
| Component C | 0.4 | 1 | 0.1667 | 0.0667 |
Your turn.
Try an answer before opening a hint. Feedback will explain the result, and you can practice again.
Interactive practice is loading. The lesson, worked example, and printable questions remain available below.
View printable practice and solutions
Use these questions when studying without interactive controls. Opening a written solution does not earn progress.
With old coefficient 0.4, signal 0.8, and α = 0.5, what is the blended new coefficient?
Read the solution
The new coefficient is 0.5 × 0.4 + 0.5 × 0.8 = 0.6.
A teaching query uses coefficients (1, 2) and features (3, 4). What is their dot product?
Read the solution
The dot product is 3 + 8 = 11.
What changes when an additive learning rule is replaced by a blend with α between zero and one?
- The old coefficient is scaled before the new signal is added.
- All old data become losslessly recoverable.
- The feature count becomes infinite.
- The learning signal no longer matters.
Read the solution
The blend scales the old coefficient, changing accumulation and forgetting.
Which evaluation directly tests whether the sketch is useful as retrieval memory?
- The number of decimal digits in a coefficient.
- The animation’s frame rate.
- Accuracy and interference on a specified set of stored and queried items.
- Its use of the word DNA.
Read the solution
A retrieval evaluation tests the behavior the representation is intended to support.
Follow the original record
Use the definitions you just practiced to read the project’s own notation, assumptions, and evidence.
The ideas to keep
- Separate encoding, updating, and querying.
- Preserve historical rule variants.
- Evaluate retrieval instead of inferring it from representation names.
Reading completion is separate from your practice evidence.
Progress is kept in this browser. How learning works