Searcharxiv⌕ Search

arXiv subjects

Karthik Iyer

Publications and source records attributed to Karthik Iyer.

8 recordsLinked to original sources

Forecasting Retail Sales at Every Grain: Trees, Foundation Models, Transformers and World Models Compared on Four Public Data Sets

A retailer makes decisions from the same demand at different grains: store orders by item and day, promotions by item and week, category reviews and supplier terms. This paper asks whether one forecast, built at the finest grain and summed to every coarser grain, can serve them all, and which model does that best. It compares LightGBM, pretrained foundation models (TimesFM, Chronos-Bolt, Timestral), the Temporal Fusion Transformer and a world model that carries each item's state forward under the price and promotion plan, on M5, six Dominick's categories, Breakfast at the Frat and Corporacion Favorita, with the world model and LightGBM tuned by a common search. When to use the average: the average of the tuned world model and LightGBM, summed to every grain, has the lowest error in 22 of 27 grain and data set cases and is never the least accurate on the final held-out periods. When to use the world model: on existing items kept out of training it loses at most 0.040 in scaled error, against up to 0.311 for LightGBM, and on lines with short history it is clearly more accurate on six of nine data sets. When to use LightGBM: on heavily promoted lines, fast sellers and lines with long history (six of nine data sets each) and in price-cut weeks. Inconclusive: lines never promoted and slow lines are close to even, and 52 weeks ahead the average is the most accurate on five of seven data sets, with the world model ahead of LightGBM on two of seven. The findings hold for these data sets and the price and promotion levers; anyone adopting them should rerun every comparison on their own data.

cs.LG↗

Testing World Models on Actions They Have Never Seen: Held-Out Plans, Forecast Accuracy and Price Rules in Retail

A world model is trusted to answer "what if": how its world responds to an action it has never seen. We give a test of that trust for any action-conditioned world model: actions held out of training by design and scored against a known answer; whether forecast accuracy predicts accuracy on those actions; and the same structure given to simpler models. We run it in retail, where the action is a price and promotion plan, across 2,853 fitted models. First, forecast accuracy is a fair guide between kinds of model (rank correlation with error on new plans 0.32 to 0.66) and a poor guide between versions of one world model: where merchants plan into high-demand weeks the link reverses, down to -0.69. Second, training the world model to learn the effect of actions (JEPA pretraining, a latent consistency check, adaptation at the plan date) leaves its error on new plans unchanged. Third, in generated worlds the recorded regular price and three price rules improve it. Given the same rules, a log-linear model is best when its curve can match the truth; in a world whose curve lies outside the rules, the world model with the rules is ahead of the log-linear model with the rules in 36 of 40 worlds, and does as well without its recurrent cell. On real data (Breakfast at the Frat, The Complete Journey and M5) the rules lower the world model's error on pure price cuts deeper than any it saw (0.63 to 0.56 on the Frat; 0.45 to 0.44, within chance, on The Complete Journey), but not on lasting price moves. Fourth, a small share of random test plans helps every model a little. In a generated dosing world, a stress test in a second domain, the first finding repeats, and the built-up drug level lowers the world model's error (2.86 to 1.91) though it still predicts only 0.12 to 0.32 of the true gain. The test carries to any domain where a world model is asked about actions outside its history.

cs.LG↗

M-JEPA: Predictive Self-Supervised Learning for Molecular Graphs with Scaffold-Shift Evaluation on Tox21

Self-supervised molecular representation learning is increasingly used to improve transfer on small, label-limited property prediction tasks, yet commonly used contrastive objectives can be sensitive to view construction and are often evaluated with protocols that report only discrimination. We adapt Joint Embedding Predictive Architectures (JEPAs) to molecular graphs by masking chemically coherent connected subgraphs and training a context encoder to regress the masked target embedding produced by an exponential-moving-average (EMA) teacher. To evaluate predictive self-supervision under realistic distribution shift, we introduce a leakage-resistant, compute-matched three-phase protocol: (i) an 80-run objective screening on ESOL with fixed compute budgets, (ii) JEPA-only refinement of masking and augmentation choices, and (iii) per-assay Tox21 evaluation under Bemis-Murcko scaffold splits with joint discrimination-reliability analysis and attribution diagnostics. Under matched compute, M-JEPA achieves substantially lower ESOL proxy RMSE than a compute-matched InfoNCE baseline (mean RMSE 2.12 vs 3.75; mean paired dRMSE 1.61; 95% bootstrap CI [1.52, 1.72]; Wilcoxon p=2.4e-4) and typically completes runs faster (~15-25 min vs ~35-45 min on a single GPU under identical hardware), albeit with higher runtime variance. On Tox21, improvements in ROC-AUC/PR-AUC are assay-dependent and do not uniformly imply better calibration; reliability diagrams and Brier/ECE metrics reveal endpoints where fine-tuning improves ranking but leaves probabilities miscalibrated.

cs.LG↗

Orbital dynamics and precession in magnetized Kerr spacetime

We study the orbital structure and precession dynamics of neutral test particles in the magnetized Kerr black hole (MKBH) spacetime-an exact electrovacuum solution of the Einstein-Maxwell equations that self-consistently incorporates the curvature effects of an external magnetic field. This geometry allows a unified treatment of gravitational and magnetic influences across weak to ultra-strong regimes. The analysis reveals a critical magnetic field strength above which no circular geodesics, timelike or null, can exist, establishing an upper magnetic bound for orbital motion. For subcritical fields, the photon circular orbit admits two real roots, the outer of which defines an outermost stable circular orbit (OSCO), complementing the conventional innermost stable circular orbit (ISCO) and confining stable motion within a finite radial domain. Exact expressions for the orbital, radial, and vertical epicyclic frequencies, and their associated precession rates, show substantial deviations from Kerr behavior, including a magnetically induced reversal of periastron precession within a finite radial range. For astrophysically relevant magnetic field strengths, the retrograde precession could be observable at large radii around astrophysical BHs, offering a potential diagnostic of large-scale magnetization. These findings highlight the geometric influence of magnetic curvature on strong-field dynamics, providing a self-consistent framework to interpret quasi-periodic oscillation phenomenology and potential magnetic imprints in precision timing observations of compact objects.

gr-qc↗

Semantic IDs for Recommender Systems at Snapchat: Use Cases, Technical Challenges, and Design Choices

Effective item identifiers (IDs) are an important component for recommender systems (RecSys) in practice, and are commonly adopted in many use cases such as retrieval and ranking. IDs can encode collaborative filtering signals within training data, such that RecSys models can extrapolate during the inference and personalize the prediction based on users' behavioral histories. Recently, Semantic IDs (SIDs) have become a trending paradigm for RecSys. In comparison to the conventional atomic ID, an SID is an ordered list of codes, derived from tokenizers such as residual quantization, applied to semantic representations commonly extracted from foundation models or collaborative signals. SIDs have drastically smaller cardinality than the atomic counterpart, and induce semantic clustering in the ID space. At Snapchat, we apply SIDs as auxiliary features for ranking models, and also explore SIDs as additional retrieval sources in different ML applications. In this paper, we discuss practical technical challenges we encountered while applying SIDs, experiments we have conducted, and design choices we have iterated to mitigate these challenges. Backed by promising offline results on both internal data and academic benchmarks as well as online A/B studies, SID variants have been launched in multiple production models with positive metrics impact.

cs.IR↗

Spin Precession in magnetized Kerr spacetime

We present an exact analytical investigation of spin precession for a test gyroscope in the magnetized Kerr spacetime--an exact electrovacuum solution to the Einstein-Maxwell equations. Our approach accommodates arbitrary magnetic field strengths, enabling a unified treatment across both weak and ultra-strong field regimes. The analysis reveals distinct spin precession behaviors near rotating collapsed objects, which differ characteristically between black holes and naked singularities, offering a potential observational means to differentiate them. The external magnetic field induces a nontrivial modification of the precession frequency through its interaction with the spacetime's gravitoelectromagnetic structure. In the weak-field limit, magnetic fields generally reduce the precession rate, though the effect depends sensitively on the motion and orientation of the test gyro close to the collapsed object. As a special case, we show that in the presence of magnetic fields, the spin precession frequency due to gravitomagnetic effect acquires a long-range $1/r$ (where $r$ is the distance from the central object to the test gyro) correction in contrast to the standard $1/r^3$ falloff. In addition, we obtain the exact geodetic precession (gravitoelectric effect) frequency for a gyroscope in magnetized Schwarzschild spacetime, showing that the magnetic field enhances ($\propto r^{1/2}$) geodetic precession in contrast to the standard $1/r^{5/2}$ falloff. Our results provide observationally testable predictions relevant for black holes in strong magnetic environments, including those possibly realized near magnetars or in the early universe. In particular, the strong-field behavior of spin precession could have important implications for transmuted black holes formed via collapse or mergers of magnetized progenitors in both astrophysical and cosmological contexts.

gr-qc↗

Determining rough first order perturbations of the polyharmonic operator

We show that the knowledge of Dirichlet to Neumann map for rough $A$ and $q$ in $(-Δ)^m +A\cdot D +q$ for $m \geq 2$ for a bounded domain in $\mathbb{R}^n$, $n \geq 3$ determines $A$ and $q$ uniquely. This unique identifiability is proved via construction of complex geometrical optics solutions with sufficient decay of remainder terms, by using property of products of functions in Sobolev spaces.

math.AP↗

Cloaking for a quasi-linear elliptic partial differential equation

In this article we consider cloaking for a quasi-linear elliptic partial differential equation of divergence type defined on a bounded domain in $\mathbb{R}^N$ for $N=2,3$. We show that a perfect cloak can be obtained via a singular change of variables scheme and an approximate cloak can be achieved via a regular change of variables scheme. These approximate cloaks though non-degenerate are anisotropic. We also show, within the framework of homogenization, that it is possible to get isotropic regular approximate cloaks. This work generalizes to quasi-linear settings previous work on cloaking in the context of Electrical Impedance Tomography for the conductivity equation.

math.AP↗