k-NN Add ADC and Random Rotation documentation to binary quantization (#10657)

* k-NN Add ADC and Random Rotation documentation to binary quantization

Signed-off-by: Finn Roblin <finnrobl@amazon.com>

* Apply suggestions from code review

Signed-off-by: kolchfa-aws <105444904+kolchfa-aws@users.noreply.github.com>

* Apply suggestions from code review

Signed-off-by: Nathan Bower <nbower@amazon.com>

---------

Signed-off-by: Finn Roblin <finnrobl@amazon.com>
Signed-off-by: kolchfa-aws <105444904+kolchfa-aws@users.noreply.github.com>
Signed-off-by: Nathan Bower <nbower@amazon.com>
Co-authored-by: kolchfa-aws <105444904+kolchfa-aws@users.noreply.github.com>
Co-authored-by: Nathan Bower <nbower@amazon.com>

Author: 3 people

_vector-search/optimizing-storage/binary-quantization.md

@@ -117,6 +117,56 @@ PUT my-vector-index
 ```
 {% include copy-curl.html %}
 
+## Enhancing search quality with ADC and RR
+**Introduced 3.2**
+{: .label .label-purple }
+
+If your search results have low recall, you can enhance search quality by specifying asymmetric distance computation (ADC) or random rotation (RR) in the index mapping.
+
+ADC maintains a full-precision query vector while rescaling it to have meaningful distance computations against binary-quantized document vectors. This asymmetric approach preserves more information about the query vector, boosting search quality without significant memory penalty. ADC is supported for 1-bit quantization only.
+
+RR addresses the issue of binary quantization giving equal weight to each vector dimension during the quantization process. By rotating the distribution, RR can redistribute variance (information) from high-variance dimensions to low-variance dimensions, preserving more information during the 32x compression process. RR is supported for 1-bit, 2-bit, and 4-bit quantization.
+
+For optimal performance and recall enhancement, use both ADC and RR together:
+
+```json
+PUT vector-index
+{
+  "settings" : {
+    "index": {
+      "knn": true
+    }
+  },
+  "mappings": {
+    "properties": {
+      "vector_field": {
+        "type": "knn_vector",
+        "dimension": 8,
+        "method": {
+            "name": "hnsw",
+            "engine": "faiss",
+            "space_type": "l2",
+            "parameters": {
+              "encoder": {
+                "name": "binary",
+                "parameters": {
+                  "bits": 1,
+                  "random_rotation": true,
+                  "enable_adc": true
+                }
+              }
+            }
+        }
+      }
+    }
+  }
+}
+```
+{% include copy-curl.html %}
+
+ADC and RR impact search and indexing performance, so they are disabled by default. ADC may introduce a moderate latency increase because of full-precision distance computations, while RR primarily affects indexing latency because vectors must be rotated during the process.
+{: .note}
+
 ## Search using binary quantized vectors
 
 You can perform a vector search on your index by providing a vector and specifying the number of nearest neighbors (k) to return: