Source author record

Soham Ghosh

Explore this author’s publications, collaborators and research topics. This catalog record has not been claimed by its author.

ResearcherUnclaimed source record

Applications Computation and Language math.AC

Catalog footprint

Research overview

3works

3topics

4close collaborators

Actions

Is this your work?

Open graph Browse papers

preprint2026arXiv

Ministral 3

We introduce the Ministral 3 series, a family of parameter-efficient dense language models designed for compute and memory constrained applications, available in three model sizes: 3B, 8B, and 14B parameters. For each model size, we release three variants: a pretrained base model for general-purpose use, an instruction finetuned, and a reasoning model for complex problem-solving. In addition, we present our recipe to derive the Ministral 3 models through Cascade Distillation, an iterative pruning and continued training with distillation technique. Each model comes with image understanding capabilities, all under the Apache 2.0 license.

preprint2021arXiv

The Zariski covering number for vector spaces and modules

Given a $K$ -vector space $V$ , let $σ(V,K)$ denote the covering number, i.e. the smallest (cardinal) number of proper subspaces whose union covers $V$ . Analogously, define $σ(M,R)$ for a module $M$ over a unital commutative ring $R$ ; this includes the covering numbers of Abelian groups, which are extensively studied in the literature. Recently, Khare-Tikaradze [Comm. Algebra, in press] showed for several classes of rings $R$ and $R$ -modules $M$ that $σ(M,R)=\min_{\mathfrak{m}\in S_M} |R/\mathfrak{m}| + 1$ , where $S_M$ is the set of maximal ideals $\mathfrak{m}$ such that $\dim_{R/\mathfrak{m}}(M/\mathfrak{m}M)\geq 2$ . (That $σ(M,R)\leq\min_{\mathfrak{m}\in S_M}|R/\mathfrak{m}|+1$ is straightforward.) Our first main result extends this equality to all $R$ -modules with small Jacobson radical and finite dual Goldie dimension. We next introduce a topological counterpart for finitely generated $R$ -modules $M$ over rings $R$ , whose 'some' residue fields are infinite, which we call the Zariski covering number $σ_τ(M,R)$ . To do so, we first define the "induced Zariski topology" $τ$ on $M$ , and now define $σ_τ(M,R)$ to be the smallest (cardinal) number of proper $τ$ -closed subsets of $M$ whose union covers $M$ . We first show that our choice of topology implies that $σ_τ(M,R)\leqσ(M,R)$ , the covering number. We then show our next main result: $σ_τ(M,R)=\min_{\mathfrak{m}\in S_M} |R/\mathfrak{m}|+1$ , for all finitely generated $R$ -modules $M$ for which (a) the dual Goldie dimension is finite, and (b) $\mathfrak{m}\notin S_M$ whenever $R/\mathfrak{m}$ is finite. As a corollary, this alternately recovers the above formula for the covering number $σ(M,R)$ of the aforementioned finitely generated modules. We also extend these topological studies to general finitely generated $R$ -modules, using the notion of $κ$ -Baire spaces.

preprint2020arXiv

A new generalized newsvendor model with random demand

Newsvendor problem is an extensively researched topic in inventory management. In this class of inventory problems, shortage and excess costs are considered to be proportional to the quantity lost. But, for critical goods or commodities, inventory decision is a typical example where, excess or shortage may lead to greater losses than merely the total cost. Such a problem has not been discussed much in the literature. Moreover, majority of the existing literature assumes the demand distribution to be completely known. In this paper, we propose a generalization of the newsvendor problem for critical goods or commodities with higher shortage or excess losses but of same degree. We also assume that, the parameters of the demand distribution are unknown. We also discuss different estimators of the optimal order quantity based on a random sample of demand. In particular, we provide different estimators based on (i) full sample and (ii) broken sample data (i.e with single order statistic). We also report comparison of the estimators using simulated bias and mean square error (MSE).

Inspect adjacent papers, topics, institutions and collaborators without losing the researcher page.

Building this map preview

BZPEER is loading the nearby papers, people, topics and institutions for this page.