Latent space configuration for neural network finetuning, fast ranking, and embedding dimensionality reduction

Nikita Gabdullin; Anton Raskovalov

cs.CVOct 21, 2026

Latent space configuration for neural network finetuning, fast ranking, and embedding dimensionality reduction

Authors: Nikita Gabdullin, Anton Raskovalov

Organizations: Joint Stock "Research and production company "Kryptonite"

Abstract

Finding K most similar elements for a query in a large gallery of neural network (NN) embeddings is essential in various areas of computer science and industrial applications of artificial intelligence. This paper proposes search algorithms for structured latent spaces (LS) of NNs which allow to speed up top K item retrieval owing to O(1) complexity small candidate gallery subset search. ArcFace-pretrained IResNet50 models are finetuned using Latent Space Configuration (LSC) methodology to obtain new distribution of embeddings with desired geometric properties. Applying the proposed algorithms to embeddings obtained from finetuned models leads to several times speed-up over baseline exhaustive search. Specifically, we report up to 3 times speed-up on GPU and 40% speed-up on CPU compared to the baseline method GPU speed. These results are obtained for the original and reduced numbers of LS dimensions. The results show that LS dimensionality reduction allows to further accelerate the search and even improve rank-N metrics.

Figures & tables

Figure 1: Pretrained model finetuning with LSC. Black arrows indicate operations done once before finetuning. Blue arrows indicate the parameter update loop.

Figure 2: Example of finding $V_{n}^{11}$ vectors close to some mean embedding vector when the closest $V_{n}^{11}$ vector is already assigned an id .

Figure 3: Pretrained model finetuning while reducing the number of LS dimensions. Here fc is fully-connected projection layer initialized with the same matrix Q , which is used to project embeddings from n orig into n red . Black arrows indicate operations done once before finetuning. Blue arrows show the parameter update loop.

Figure 4: Curves of loss (left) and id prediction accuracy (right) during finetuning.

Method	Device	Time, s	Pretrained	Exp I	Exp II	Exp III
			R1/R10	R1/R10	R1/R10	R1/R10
Queries in train set	-	-	Yes	Yes	No	No
Augmentation	-	-	Yes	No	No	Yes
exh. search	GPU	17.5	0.884/0.928	0.944/0.964	0.801/0.83	0.83/0.86
exh. search	CPU	336.4	0.884/0.928	0.944/0.964	0.801/0.83	0.83/0.86
Alg. 1	CPU	12.5	0.266/0.27	0.942/0.958	0.798/0.81	0.818/0.835

Table 1: Gallery ranking metrics and time required for their calculation for n orig =512.

Method	Device	Time, s	Pretrained	Exp I	Exp II	Exp III
			R1/R10	R1/R10	R1/R10	R1/R10
Queries in train set	-	-	Yes	Yes	No	No
Augmentation	-	-	Yes	No	No	Yes
exh. search	GPU	12.9	0.87/0.914	0.97/0.99	0.82/0.854	0.846/0.875
exh. search	CPU	280	0.87/0.914	0.97/0.99	0.82/0.854	0.846/0.875
Alg. 1	CPU	7.1	0.266/0.27	0.97/0.99	0.786/0.796	0.825/0.839

Table 2: Gallery ranking metrics and time required for their calculation for n red =160.

Representation Learning

Representation learning develops data transformations that expose useful structure for prediction, transfer, generation, and control.

See all 798 papers

Panorama: Fast-Track Nearest Neighbors
Oct 1, 2025Vansh Ramani, Alexis Schlomer, Akash Nayar +3Product Quantization Approximate Nearest Neighbor Search
IIT Delhi & CMU & UCPH · Carnegie Mellon University · Indian Institute of Technology Delhi +1
Clark Hash: Stateless Sparse Johnson-Lindenstrauss Quantization for Neural Embeddings
May 27, 2026Stanislav Kirdey, Clark Labs IncEmbedding Compression Neural Network Quantization
Clark Labs Inc.
RSLM: Training-Free Vector Quantization for Approximate Nearest Neighbor Search
Aug 31, 2026Rastislav Lenhardt, Teodora Dobos, Thomas Vecchiato +2Rotation-Based Quantization Vector Quantization
Google Zurich, Switzerland · Technical University of Munich Munich, Germany · University of Copenhagen Copenhagen, Denmark +1

Oct 1, 2025J/K move · Enter open · S save

Panorama: Fast-Track Nearest Neighbors

Vansh Ramani, Alexis Schlomer, Akash Nayar, Sayan Ranu+2

IIT Delhi & CMU & UCPH · Carnegie Mellon University · Indian Institute of Technology Delhi+1

Product Quantization Approximate Nearest Neighbor Search

Open paper AskPDF arXiv

Approximate Nearest-Neighbor Search (ANNS) pipelines for high-dimensional neural embeddings spend the bulk of their query time in candidate verification, making it the primary bottleneck in the search process. In this paper, we present PANORAMA, a state-of-the-art refinement technique that accelerates verification by exploiting the inherent spectral decay of these embeddings. Using PCA to compact signal energy, PANORAMA evaluates candidate distances incrementally, computing at each step a strict lower bound on the full-vector distance and dynamically pruning candidates the moment this bound exceeds the running k-th nearest neighbor distance. While PCA's concentration of variance facilitates pruning, it breaks the uniform-variance assumption required by Product Quantization (PQ); we resolve this with a variance-shaping step that redistributes energy across subvectors, rendering accretive refinement compatible with quantized indexes. Optimized for modern memory hierarchies via vectorized bulk-pruning and cache-conscious data layouts, PANORAMA has been upstreamed into the FAISS library across major index families (IVFPQ/Flat, HNSW, and Refine). PANORAMA achieves higher QPS at any target recall with a cost that provably scales inversely with dataset spectral decay, delivering end-to-end speedups of up to 28.9x and outperforming probabilistic methods across all recall bands.

Abstract

Representation Learning

Explore similar work

Panorama: Fast-Track Nearest Neighbors

Clark Hash: Stateless Sparse Johnson-Lindenstrauss Quantization for Neural Embeddings

RSLM: Training-Free Vector Quantization for Approximate Nearest Neighbor Search