arXiv subjects

Ashish Rastogi

Publications and source records attributed to Ashish Rastogi.

6 recordsLinked to original sources

arXiv2026

Multimedia Asset Personalization via Multimodal Embeddings at Netflix

Personalized promotional assets, namely artwork images and video preview clips, are critical to content discovery on Netflix. Traditional models for asset selection rely on ID-based interaction history, leaving them blind to asset content and unable to serve newly launched titles and assets. We describe how multimodal embeddings reshaped production systems at Netflix and report transferable lessons for practitioners adopting foundation-model embeddings into recommender systems. First, pretrained image embeddings unlock cross-title, cross-canvas knowledge transfer. Augmenting a two-tower model with CLIP image embeddings lets a single model serve all five Netflix artwork canvas types, replacing five separately trained per-canvas models and substantially improving cold-start performance. A lightweight extension reuses CLIP's joint text-image space to make artwork personalization query-aware in search. Second, multimodality decisively beats any single modality for video preview personalization. We describe MediaFM, our in-house tri-modal foundation model trained on a large-scale corpus of shots from the Netflix show catalog, fusing visual (SeqCLIP), audio (wav2vec 2.0), and timed-text signals; adopted for video preview personalization, it outperforms strong visual-only baselines both offline and in online A/B tests. Third, a simple offline proxy task whose performance correlates with online outcomes can accelerate the experimentation and productization cycle. Predicting the popularity-based winner from embeddings alone ranks embedding models and versions, pruning the choice space before any end-to-end integration or A/B test; it now gates every new MediaFM checkpoint. We also share the production engineering decisions (shared embedding infrastructure, low-latency serving, cheap screening) that made these deployments viable, along with the design tradeoffs and failure modes we encountered.

cs.IR↗

arXiv2026

GenRec: An LLM-Backed Recommendation Ranker at Netflix

Large language models (LLMs) are reshaping recommender systems by enabling richer modeling of users, content, and context directly in natural language. At Netflix, we are exploring this direction through GenRec, an LLM-backed recommendation ranker built on top of an in-house foundational LLM. GenRec follows a two-phase framework: Phase 1 adapts an open-source LLM to Netflix data, developing deep understanding of the catalog and member behavior while balancing capabilities such as content understanding and instruction following. Phase 2 post-trains this foundation model with recommendation-ranking specific data, labels, and reward signals, aiming to align the ranker with business requirements and long-term member satisfaction. This paper focuses on Phase 2 and the transition from a traditional discriminative ranker with thousands of engineered features to an LLM-backed ranker driven by verbalized user histories and context. We describe our design for input verbalization and context engineering, post-training data construction, reward integration, model architecture, and a cost-constrained serving design based on a prefill-only inference approach. We report results from a large-scale A/B test comparing GenRec against the current production ranker model, where we show that a GenRec model trained with substantially fewer Phase-2 labeled training examples and input signals can achieve statistically significant gains in offline and online metrics. We discuss how LLM-backed recommenders could shift the recommendation paradigm: from feature engineering to context engineering, and from bespoke architectures to shared foundation backbones. We also outline practical lessons for serving such systems under real-world resource constraints.

cs.IR↗

arXiv2012

Tatonnement in Ongoing Markets of Complementary Goods

This paper continues the study, initiated by Cole and Fleischer, of the behavior of a tatonnement price update rule in Ongoing Fisher Markets. The prior work showed fast convergence toward an equilibrium when the goods satisfied the weak gross substitutes property and had bounded demand and income elasticities. The current work shows that fast convergence also occurs for the following types of markets: - All pairs of goods are complements to each other, and - the demand and income elasticities are suitably bounded. In particular, these conditions hold when all buyers in the market are equipped with CES utilities, where all the parameters $ρ$, one per buyer, satisfy $-1 < ρ\le 0$. In addition, we extend the above result to markets in which a mixture of complements and substitutes occur. This includes characterizing a class of nested CES utilities for which fast convergence holds. An interesting technical contribution, which may be of independent interest, is an amortized analysis for handling asynchronous events in settings in which there are a mix of continuous changes and discrete events.

cs.GT↗

arXiv2010

Discrete Price Updates Yield Fast Convergence in Ongoing Markets with Finite Warehouses

This paper shows that in suitable markets, even with out-of-equilibrium trade allowed, a simple price update rule leads to rapid convergence toward the equilibrium. In particular, this paper considers a Fisher market repeated over an unbounded number of time steps, with the addition of finite sized warehouses to enable non-equilibrium trade. The main result is that suitable tatonnement style price updates lead to convergence in a significant subset of markets satisfying the Weak Gross Substitutes property. Throughout this process the warehouse are always able to store or meet demand imbalances (the needed capacity depends on the initial imbalances). Finally, our price update rule is robust in a variety of regards: 1. The updates for each good depend only on information about that good (its current price, its excess demand since its last update) and occur asynchronously from updates to other prices. 2. The process is resilient to error in the excess demand data. 3. Likewise, the process is resilient to discreteness, i.e. a limit to divisibility, both of goods and money.

cs.GT↗

arXiv2009

Stability Analysis and Learning Bounds for Transductive Regression Algorithms

This paper uses the notion of algorithmic stability to derive novel generalization bounds for several families of transductive regression algorithms, both by using convexity and closed-form solutions. Our analysis helps compare the stability of these algorithms. It also shows that a number of widely used transductive regression algorithms are in fact unstable. Finally, it reports the results of experiments with local transductive regression demonstrating the benefit of our stability bounds for model selection, for one of the algorithms, in particular for determining the radius of the local neighborhood used by the algorithm.

cs.LG↗

arXiv2008

General Algorithms for Testing the Ambiguity of Finite Automata

This paper presents efficient algorithms for testing the finite, polynomial, and exponential ambiguity of finite automata with $ε$-transitions. It gives an algorithm for testing the exponential ambiguity of an automaton $A$ in time $O(|A|_E^2)$, and finite or polynomial ambiguity in time $O(|A|_E^3)$. These complexities significantly improve over the previous best complexities given for the same problem. Furthermore, the algorithms presented are simple and are based on a general algorithm for the composition or intersection of automata. We also give an algorithm to determine the degree of polynomial ambiguity of a finite automaton $A$ that is polynomially ambiguous in time $O(|A|_E^3)$. Finally, we present an application of our algorithms to an approximate computation of the entropy of a probabilistic automaton.

cs.CC↗