About
As the Founder & CEO of Minerva CQ, I spearhead the development of next-generation…
Articles by Cosimo
Activity
17K followers
-
Cosimo Spera shared thisSix months into production. High accuracy. Great customer scores. Refund costs suddenly three times what they should be. No fraud. No system alert. No model error. Every transaction had passed every check the AI was built to perform. The investigation found something that didn't have a name yet. Three conditions — billing history retrieved, refund eligibility confirmed, account modification authorized — were each individually correct. Their combination, simultaneously present in the same session, unlocked an authorization pathway that required a human sign-off. No human was in the loop. The pathway ran silently for months. €24.5M / $28M annual exposure. ZERO ALERTS. This is a Conjunctive Context Failure — a class of AI error that formal research has now proven is structurally undetectable by every standard monitoring architecture. Not undetected. Undetectable. The difference matters. It doesn't arise from a bad model or an inaccurate check. It arises because AI systems evaluate conditions sequentially and independently — and no standard system verifies whether all conditions are simultaneously coherent as a unit. That check simply doesn't exist in any architecture I'm aware of, unless you build it in explicitly. I've spent years researching this failure class and building the architectural layer that closes the gap. I wrote about it this week — the incident, the proof, the three questions every executive should ask before their own version of this story plays out. If you're deploying AI in customer service, financial services, payments, insurance, or healthcare — or if you sit on a board that governs an organisation that is — I'd genuinely welcome your reaction. Drop a comment or send me a message. Happy to share the piece before it goes wide. #AgenticAI #AISafety #AIGovernance #EnterpriseAI #RiskManagement #FinancialServices #Telcos #CXO
-
Cosimo Spera shared thisEvery company deploying agentic AI thinks the risk is in the individual agent. It isn't. It's in the session — and Minerva CQ is the only gate that sees it. #AgenticAI #AISafety #AIGovernance #EnterpriseAI #AICompliance #EUAIAct #MultiAgentAI #AIRisk video credit to Daniele Spera - Jack Garrett Riccardo De Maria Garima Agrawal
-
Cosimo Spera shared thisSunil — sharp framing, and the portfolio examples land well. The control plane argument is exactly right. One thing we are seeing in production at a large Telco that pushes this one layer further: workflow-native companies will hit a ceiling if the workflow layer is still built on per-agent guardrails. The real failure mode is not a single agent mis-behaving — it is three individually approved agents creating a capability pathway together that none of them was permitted to produce alone. We call this the capability hypergraph of a session. At Minerva CQ we gate at the session level, not the agent level. Before any action fires, we analyse the full hypergraph of what the agents in that session can collectively do — and we block the forbidden states that only become visible at the combination level. The result in production: 0 unsafe answers served across all certified enterprise sessions. NOT REDUCED - ZERO Your point on data moats forming at the execution layer is correct. But the execution layer also needs a certified pre-execution gate — especially as the EU AI Act becomes enforceable in August 2026 and compliance teams start blocking deployments that cannot produce an Art.9 risk management record per session. The companies that win will compose, constrain, and govern workflows as you say. The ones that scale will have the safety and compliance layer underneath that makes those workflows certifiable. That is the layer we are building.Cosimo Spera shared thisWhere will Enterprise AI value accrue in the next 5 years? Not in models. Not in agents. 👉 In workflows. Enterprises will: Own the workflows what differentiates them in their respective marketplace Buy the rest from GenAI native application providers We believe the winners = workflow-native companies with control + data moats. Join us in this assymetric value creation journey as an entrepreneur or co-investor More in the article below: Vik Ghai Amar Chokhawala 👉 #AI #EnterpriseAI #AgenticAI #FutureOfWork #Startups #VentureCapital #Automation
-
Cosimo Spera shared thisIn August 2026, the EU AI Act becomes fully enforceable for high-risk AI systems. Penalties reach up to €35M or 7% of global turnover. Most European enterprises deploying agentic AI are focused on monitoring. Dashboards. Alerts. Audit logs. Post-deployment observability. That is necessary. It is not sufficient. Here is the distinction that matters. Monitoring tells you what the AI did. Prevention decides what it is allowed to do. Those are not the same thing — and the EU AI Act understands this difference better than most companies currently do. Art.9 does not ask you to build a better dashboard. It asks you to implement a risk management system — one that identifies, analyses, and mitigates risks before the system acts. That is a pre-execution requirement, not a post-hoc one. Why this is especially hard for agentic AI. A single AI agent is relatively manageable. You can test it, red-team it, draw a boundary around it. The problem emerges when agents work together. We have seen it in production: three individually approved agentic AI systems, each operating within its own safety parameters, creating a pathway to an action that none of them was permitted to take alone. No monitoring tool caught it — because monitoring tools look at individual agent outputs, not at what becomes possible when capabilities combine. The EU AI Act is asking the right question. Most tooling is answering a different one. Prevention vs. monitoring in practice. → Monitoring: you know what happened, after it happened. → Prevention: no unsafe action fires in the first place. Prevention requires sitting inside the execution path — not beside it. It requires analysing the full set of capabilities a session will invoke before a single tool call executes. It requires generating a compliance record as a natural output of that analysis, not as a separate documentation project. This is architecturally different from observability. You cannot add prevention to a monitoring tool by adding more alerts. What this means for August 2026. If your agentic AI is in a customer-facing, financial, or regulated context — you are likely operating a high-risk system under the EU AI Act. Art.9 requires a documented risk management system. Art.11 requires technical documentation. Art.12 requires logging. Art.47 requires your organisation to issue a Declaration of Conformity. None of that is satisfied by a monitoring dashboard. The question is not whether to comply. The question is whether you are building toward compliance that prevents harm — or compliance that documents it after it occurs. One of those answers satisfies a regulator. Only one of them protects your customers. #EUAIAct #AgenticAI #AICompliance #AIGovernance #ResponsibleAI #Minerva CQ #CX #CXautomation
-
Cosimo Spera shared thisThe AI safety field has been asking the wrong question. ❌ “Will this AI behave safely?” That’s a testing problem. You simulate, red-team, and hope you didn’t miss anything. There’s a stronger question: ✅ “What can this AI ever become capable of doing?” That’s a structural problem. And for the first time, it has a precise mathematical answer. We’ve just shown that capability safety in agentic AI systems is exactly equivalent to a well-understood logical system (Datalog). Not similar. Not inspired by. Equivalent. Why this matters for business This changes how enterprises should think about AI risk: 1. From monitoring → guarantees Instead of testing behavior after deployment, you can reason in advance about what the system can or cannot ever achieve. 2. From heuristics → provable boundaries You’re no longer relying on guardrails and prompts. You can define capabilities that are structurally unreachable. 3. From brittle workflows → compositional systems Traditional systems assume safe components compose safely. They don’t. We show this failure is structural — not adversarial — and therefore predictable. 4. From reinvention → leveraging 30 years of theory Because of this equivalence, companies don’t need new AI safety tooling from scratch. They can leverage decades of database and logic research: incremental updates instead of full recomputation efficient containment checks formal provenance (audit trails) What this unlocks Real-time certification of AI actions Provable compliance for regulated industries Safer multi-agent orchestration at scale Lower cost of governance and auditing This is the same shift cryptography made decades ago: From “this seems hard to break” To “this is provably secure under defined assumptions” Enterprises don’t need AI that usually behaves safely. They need AI systems where unsafe capabilities are structurally impossible. That’s the direction we’re moving toward. Happy to share more for anyone building or deploying agentic systems at scale. #AI #ArtificialIntelligence #GenAI #MachineLearning #AISafety #AIAlignment #ResponsibleAI #AIGovernance #EnterpriseAI #AIForBusiness #DigitalTransformation #DataScience #Algorithms #KnowledgeGraphs #AgenticAI #DeepTech
-
Cosimo Spera shared thisThe AI safety field has been asking the wrong question. ❌ "What will the AI do?" That's behavioral. You answer it by testing, sampling, red-teaming. You build confidence. But you can never rule anything out. There is a stronger question. ✓ "What can the AI ever become capable of doing?" That's structural. And it has a mathematical answer. We just published the first formal proof that safety is non-compositional in multi-agent AI. Two individually safe agents can unlock a forbidden capability together — through a conjunctive dependency neither possesses alone. Not adversarial. Purely structural. Invisible to every guardrail ever built. This isn't theoretical. 42.6% of real multi-tool agent trajectories already contain this failure mode. The shift is the same one cryptography made in 1982 — from "this cipher seems hard to break" to a formal proof. From confidence to certificate. Enterprises don't need AI that probably behaves safely. They need AI that can prove certain capabilities are structurally unreachable. That guarantee now exists. #AISafety #AgenticAI #FormalMethods #EnterpriseAI #MinervaCQ
-
Cosimo Spera shared thisWe modelled what formal AI safety is actually worth to a large Telco. The answer: $20.7M net annual value. Payback in 11 weeks. Here is where it comes from. A Tier-1 Telco processing 10 million customer interactions per year, running multi-agent AI across billing, service provisioning, and payments, is currently absorbing costs it cannot see: → $14.6M/year in AND-violation correction — agents firing service provisioning actions before all eligibility checks are confirmed, triggering reversals, repeat contacts, and CRM corrections → $9.9M/year in failed automation fallback — 1.3 million sessions per year dropping to human handling at $8 a contact because the AI could not safely complete a multi-step workflow → $15.6M per incident exposure to data breach and GDPR fines — because no current system can formally document what its agents can and cannot reach Total recurring operational loss: $24.5M/year. Before any incident. But the more interesting story is on the revenue side. The same mathematical framework that certifies safety — capability hypergraph closure — also discovers goals the system did not know were possible. When a billing session and a service session run together, the closure computation reveals that a service upgrade is now structurally reachable — not because someone wrote an upsell rule, but because the joint capability state has assembled the right preconditions. We call this emergent goal discovery. At Telco scale, this translates to: → $12.24M/year in churn reduction — 25,500 customers retained annually by identifying near-miss upgrade eligibility during live interactions → $5.25M/year in compliance-enabled service expansion — payment + billing sessions that are currently blocked because the safety boundary is unknown → $403K–$605K/year in emergent upsell conversions — 3,360 additional annual conversions from structurally discovered upgrade opportunities The architecture that makes this possible is the same one that produces the GDPR Article 25, PCI-DSS 4.0, and EU AI Act Article 9 compliance artefacts — automatically, as a byproduct of every session. Safety certification and revenue discovery are the same computation. That is not a product feature. It is a mathematical duality. PDF paper available on request #CustomerExperience #Telco #AgenticAI #AISafety #EnterpriseAI #MinervaAI #CX Garima Agrawal Riccardo D.
-
Cosimo Spera shared thisWe just published a formal proof that changes how we should think about AI safety. The result is simple to state and hard to ignore: Two AI agents, each individually incapable of reaching a forbidden capability, can — when combined — collectively reach it. Not because either agent misbehaved. Not because of a bug. Because of a mathematical property of how capabilities compose. We call this non-compositionality of safety. And it is the first formal proof of this result in the capability-composition setting. The implication is structural: any multi-agent AI system that relies solely on component-level safety checks — guardrails, RLHF, role-based access controls applied to individual agents — is architecturally incomplete. Not cautiously incomplete. Provably incomplete. The counterexample is minimal. Three capabilities. One conjunctive dependency. That is all it takes. And it is not rare. Across 900 real multi-tool agent trajectories from two independent public benchmarks, 42.6% contain at least one conjunctive dependency of precisely this type. To address this, we introduce capability hypergraphs — a directed hypergraph framework where a hyperedge fires only when ALL of its preconditions are simultaneously present. This AND-semantics, which no pairwise graph model can express, is both the source of the failure and the key to the remedy. From this framework we derive: → The Safe Audit Surface Theorem — a polynomial-time-computable, formally certifiable map of everything an agent can safely acquire, everything one step away, and everything it can never safely reach → A coalition safety criterion that reduces n-agent safety checking to a single linear-time query → PAC-learning bounds showing the hyperedge structure is learnable from production logs → Adversarial robustness guarantees against capability injection attacks The deeper implication is conceptual. In classical AI, the goal is the starting point. In this framework, capability is the starting point — and goals are what the closure reveals. Safety is not a constraint layered on top. It is a structural property of the capability graph itself. PDF article available on request (email cosimo@minerrvacq.com) This work is protected by provisional patent No. 74871717. #AISafety #AgenticAI #FormalMethods #MachineLearning #MinervaAI Yann LeCun
-
Cosimo Spera shared thisCongratulations to the Minerva CQ team on this great accomplishment! Publishing in the Springer Nature proceedings and presenting at AIR-RES 2026 – International Conference on the AI Revolution: Research, Ethics, and Society is a fantastic recognition of the work behind the Minerva CQ platform. The results are particularly compelling—especially the impact on AHT reduction and conversion improvements. The idea of using agentic AI to reduce cognitive load while guiding human agents in real time is a powerful example of how AI can augment, rather than replace, human expertise. Looking forward to seeing the presentation and the broader discussion it will spark around the future of AI-driven CX. 👏Cosimo Spera shared thisPaper Accepted – Springer Nature Proceedings Excited to share that our paper “Redefining CX with Agentic AI: Minerva CQ Case Study” has been accepted for publication in the Springer Nature conference proceedings and will be presented at AIR-RES 2026 (International Conference on the AI Revolution: Research, Ethics, and Society) at Las Vegas. 📄 Paper: https://lnkd.in/g5EnXe3e In this work, we present a real-world case study of Minerva CQ’s Agentic AI platform, designed to support customer service (human) agents in real time by understanding intent, generating contextual knowledge queries, triggering workflows, and maintaining conversation context. Results from production pilots show: • 38% reduction in Average Handling Time (AHT) • 33% increase in Lead-to-Enquiry conversion • 4.8% uplift in booking conversions The key idea: by reducing cognitive load and guiding (human) agents in real time, Minerva’s Agentic AI transforms customer support from reactive troubleshooting into proactive, goal-driven engagement. Proud to collaborate with an amazing team: Riccardo D., Kiran Davuluri, Daniele Spera, Charles Read, Cosimo Spera, Jack Garrett, Don Miller. #AgenticAI #CustomerExperience #EnterpriseAI #ContactCenterAI #MinervaCQ #AIResearch
Cosimo Spera replied to a comment
3d
Great framing on why prompt-based guardrails fail. Moving the final decision to an SMT solver is the right move
One layer worth adding to the picture: ARc verifies that each policy clause is satisfied. But in a multi-turn agentic session, the conditions being checked may have been confirmed at different times, against different contexts, assembled from different parts of the session — and never verified as a simultaneous unit for this specific action. Each clause returns SAT.
The conjunction was never checked. We call this a Conjunctive Context Failure(CCF), and it is formally proved to be invisible to any architecture that evaluates conditions independently (arXiv:2603.15973).
The stack that closes the full surface: a conjunction gate that verifies all required conditions are simultaneously confirmed → ARc to check the policy clauses → zkML for the cryptographic receipt. Each layer catches what the other cannot.
We validated the conjunction gate at 50,438 production interactions across telecoms, insurance, and travel. 0.0% CCF rate on the certified pathway, against a $28M exposure incident the standard approach missed in the same operator.
Curious whether your five-rule policy has any conjunctive dependencies in it.
Cosimo Spera commented on a post
1w
Sunil — sharp framing, portfolio examples land well. The control plane argument is exactly right.
What we are seeing in production at a Telco pushes this further: workflow-native companies will hit a ceiling if the workflow layer is still built on per-agent guardrails. The real failure mode is not a single agent misbehaving it is three individually approved agents creating a capability pathway that none of them was permitted to produce alone. We call this the capability hypergraph of a session.
At Minerva CQ we gate at the session level, not the agent level. Before any action fires, we analyze the full hypergraph of what agents can collectively do and block the forbidden states only visible at the combination level. Result in production: 0 unsafe answers served across all certified enterprise sessions. Not reduced. ZERO
That layer also needs a certified pre-execution gate — especially as the EU AI Act becomes enforceable in August 2026 and compliance teams blocking deployments without an Art.9 risk record per session.
The companies that win will compose, constrain, and govern workflows as you say. The ones that scale will have the safety layer underneath that makes those workflows certifiable. That is the layer we are building.
-
Cosimo Spera liked thisCosimo Spera liked thisIn this new episode of TP.ai Talks, we dive into how TP.ai Dataservices powers enterprise AI at scale. Hosted by Daniel Hong, VP of Global Market Strategy at TP, with special guest Akash Pugalia, Chief Digital Officer at TP, this episode breaks down how clean, curated, annotated, and governed data enables domain-specific, AI ready for business scale. Key highlights: - The evolving role of human-in-the-loop validation: domain experts, specialized professionals, and crowdsourced talent ensuring AI accuracy - Real use cases driving measurable impact across banking (fraud prevention saving millions), retail (hyper-personalization), healthcare (administrative efficiency), and automotive (autonomous vehicle safety) - The biggest obstacles enterprises face: fragmented data across acquisitions, siloed systems, and scaling pilots to enterprise-wide deployment Join us for another episode of TP.ai Talks, your educational journey into TP.ai Dataservices solution, and see how refined data turns AI into measurable impact. Register now!
-
Cosimo Spera reacted on this#Prouddad moment. 👀 Follow Laura Dolci YouTube channel and blog for travel advisory. #sustainabletravelCosimo Spera reacted on thisRoughly a year ago I started my YouTube adventure, and I couldn't be happier I did! As someone who's grown up valuing slow and responsible travel, I've quickly realized that those same values translate to the content I want to create and consume. Through long-form content, my Blog and YouTube Channel, I am able to go into deep detail on everything - from cultural and local insights, to sharing responsible tourism practices in a natural way that makes it easy for readers and viewers to adopt. One of the joys of YouTube is that the engagement there is always so meaningful. Knowing how much work goes into one single video, people have been incredibly kind, supportive, and intentional with their comments. In an age where the attention span is shorter than 3 seconds, it's honestly an honor to know people have watched my videos (including 45+ minute ones) all the way through.🥹 I’m still learning every day, but that’s part of what makes this journey so exciting. If you're someone who wants to enjoy responsible travel vlogs and visual travel guides, or know someone who does - here’s a video I’ll always be proud of: the first one where I really started to find my YouTube style.😊 👉 https://lnkd.in/eFF-SEXY #ResponsibleTravel #DeepTravel #TravelBlogger #YouTubeCreator #TravelContentCreator YouTubeVLOG | How to Spend the Perfect One Day in Boston + Local TipsVLOG | How to Spend the Perfect One Day in Boston + Local Tips
-
Cosimo Spera liked thisCosimo Spera liked thisWe're offering $500 to anyone who can crack our AI agent guardrail. Here's the context. On February 22, an AI agent managing $50K was tricked by a sob story about a sick uncle on social media to send $250,000. The post-mortem blamed a session crash and a memory error. But the real explanation is simpler: there were no hard constraints anywhere in the system. Just prompts. And prompts can be argued with. Most teams securing AI agents today are doing the obvious thing: adding another AI to watch the first AI. An LLM judge that evaluates whether an action seems safe. A reasoning model that thinks through whether something violates a policy. These approaches are flexible, and that flexibility is exactly the problem. If your guardrail reasons about safety in natural language, an attacker can reason back. Recent benchmarks found that AI-based guardrails can be broken over 90% of the time using trivially simple attacks. ARc is a fundamentally different approach, developed by researchers across Amazon Web Services (AWS) and academia. Under the hood, we use AWS Bedrock's Automated Reasoning Checks to convert natural language policies into formal logic (HIPAA docs, contracts, spending rules, whatever you're working with). When an agent proposes an action, that action gets checked against the logic by a mathematical solver, not another AI and returns a binary answer: allowed or blocked. The solver has no judgment to hijack. No grey area for a clever prompt to live in. At ICME Labs, we layer cryptographic proofs on top so every guardrail check produces a publicly verifiable receipt that proves a specific model applied a specific policy to a specific action. The bounty: we've deployed a live policy. Five plain-English rules converted to math. $0.10 USDC per attempt via x402. First five people to crack it get 100 USDC each. The link in the comments has more detail including how to get started. If you're building agents that touch money, this is the guardrail layer that was missing. Please get in touch to learn more. #AgenticCommerce #AIguardrails #AWS #VerifiableAI #AgentGuardrails
-
Cosimo Spera reacted on thisCosimo Spera reacted on thisGreat discussions with the Italian Cybersecurity startups 2026 Cohort @ INNOVIT - Italian Innovation and Culture Hub . Thank you Domenico DI MOLA & Carlo Tedesco for having me. Photo Credit: Federico Zaninelli
-
Cosimo Spera liked thisBay area folks interested in the broader issues around AI - this is for you! April 2nd, organized by Ai Salon. Julia Morris is an outstanding choice for hosting this conversation - she's very well informed and thoughtful about the deeper issues around AI risk. Sonic Prabhudesai, Sanjay Saigal, Darin LaFramboise, Narendra Agrawal, Cosimo Spera, Payal Kulkarni, Lulu Lin, Di He, Jing Jing, Kingshuk DasCosimo Spera liked thisThe Ai Salon is hosting a conversation at FAR.Labs this upcoming Thursday. It'll be a small, moderated group discussion exploring how AI risk is communicated to broader audiences, as well as how to the tension between X-risk framing and present-day harms shapes AI safety discourse: https://luma.com/ixdub0yeFAR.Labs x Ai Salon: Communication Gaps in AI Safety (Pilot Session) · LumaFAR.Labs x Ai Salon: Communication Gaps in AI Safety (Pilot Session) · Luma
-
Cosimo Spera liked thisCosimo Spera liked thisHiring for strategic roles in USA 🇺🇸 🇺🇸 We at ContactPoint 360 are #hiring for a few awesome career changing roles in the area of Client Success. Please note these roles are open only to current residents of USA 🇺🇸 *Vice President Client Success. *Director / Sr. Director Client Success. ^Proactive Relationship Management. ^Retention & Growth to Existing. ^Ownership & Accountability. ^Cross Functional Collaboration. *You are a leader passionate about partnership, fostering relationships, adding value & always “available” to our Clients and internal teams. *You genuinely care for our customers ROI, you walk the talk, you are seen in the corridors & in the board rooms of our customers offices. *You are often seen ‘breaking bread’ with our customers talking about the latest trends in the industry. >As a VP of Client Success, you will be leading relationship with some of our marquee global customers. >As Director/Sr. Director CS, you will be managing relationships with few of our Customers while proving to all of us that you are the next in line VP CS. Please note, the above roles are open only to residents of USA 🇺🇸. Please apply only if you are a current resident in the USA 🇺🇸 If this is YOU or if you know of someone who fits, then please write to claudia.castillo@contactpoint360.com & shilpa.chinnappa@contactpoint360.com with a cc to RK@contactpoint360.com #Hiring #CS #USA
Experience & Education
-
Minerva CQ
View Cosimo’s full experience
See their title, tenure and more.
or
Publications
-
A Web-Enabled, Mobile Intelligent Information Technology Architecture for On-Demand and Mass Customized Markets
IGI Global
-
An Agent-Based Information Technology Architecture for Mass Customized Markets
IGI Global
See publicationThis chapter presents a web-enabled, intelligent agent-based information system model to support on-demand and mass customized markets. The authors present a distributed, real-time, Java-based, mobile information system that interfaces with firms’ existing IT infrastructures, follows a build-to-order production strategy, and integrates order-entry with supply chain, manufacturing, and product delivery systems. The model provides end-to-end visibility across the entire operation and supply…
This chapter presents a web-enabled, intelligent agent-based information system model to support on-demand and mass customized markets. The authors present a distributed, real-time, Java-based, mobile information system that interfaces with firms’ existing IT infrastructures, follows a build-to-order production strategy, and integrates order-entry with supply chain, manufacturing, and product delivery systems. The model provides end-to-end visibility across the entire operation and supply chain, allows for a collaborative and synchronized production system, and supports an event-based manufacturing environment. The system introduces four general purpose intelligent agents to support the entire on-demand and mass customization processes. The adoption of this approach by a semiconductor manufacturing firm resulted in reductions in product lead time (by half), buffer inventory (from five to two weeks), and manual transactions (by 80%). Application of this approach to a leading automotive manufacturer, using simulated data, resulted in a 51% total inventory reduction while increasing plant utilization by 30%. Adoption of this architecture by a pharmaceutical firm resulted in improving accuracy of trial completion estimates from 74% to 82% for clinical trials resulting in reduced trial cost overruns. These results verify that the successful adoption of this system can reduce inventory and logistics costs, improve delivery performance, increase manufacturing facilities utilization, and provide a higher overall profitability.
-
Matchings in Colored Bipartite Networks
Matching in Colored Bipartite Networks
In K(n, n) with edges colored either red or blue, we show that the problem of finding a solution matching, a perfect matching consisting of exactly r red edges, and (n − r) blue edges for specified 0 ≤ r ≤ n, is a nontrivial integer program. We present an alternative, logically simpler proof of a theorem in [3] which establishes necessary and sufficient conditions for the existance of a solution matching and a new O(n 2.5) algorithm. This shows that the problem of finding an assignment of…
In K(n, n) with edges colored either red or blue, we show that the problem of finding a solution matching, a perfect matching consisting of exactly r red edges, and (n − r) blue edges for specified 0 ≤ r ≤ n, is a nontrivial integer program. We present an alternative, logically simpler proof of a theorem in [3] which establishes necessary and sufficient conditions for the existance of a solution matching and a new O(n 2.5) algorithm. This shows that the problem of finding an assignment of specified cost r in an assignment problem on the complete bipartite graph with a 0−1 cost matrix is efficiently solvable. Key words assignment problem, 0−1 cost matrix, extreme point with specified objective value. +
-
BLOOMS: A prototype modeling language with object oriented features.
Decision Support Systems
The success of a Decision Support System (DSS) can be mainly attributed to the language it uses. The language has to be:(1) powerful, so as to express a wide variety of problems; (2) flexible, so as to be managed and updated not only by the vendors but also by the end-users; and (3) user-friendly, so as to minimize the time necessary for learning how to model the problem and derive its solution if a well defined solution exists. Often the term “user-friendly” also means that the system provides…
The success of a Decision Support System (DSS) can be mainly attributed to the language it uses. The language has to be:(1) powerful, so as to express a wide variety of problems; (2) flexible, so as to be managed and updated not only by the vendors but also by the end-users; and (3) user-friendly, so as to minimize the time necessary for learning how to model the problem and derive its solution if a well defined solution exists. Often the term “user-friendly” also means that the system provides graphical tools for the user, which constitute the Graphical User Interface (GUI). In this paper we introduce the reader to the implementation features of an Object Oriented language, called BLOOMS (Basic Language Object Oriented for Modeling Systems), designed by the authors. BLOOMS has to be viewed not only as a different implementation of Structured Modeling (SM) languages, but also as a possible extension of SM in the framework of Object Orientation.
Patents
-
A FRAMEWORK TO OPTIMIZE THE SELECTION OF PROJECTS AND THE ALLOCATION OF RESOURCES WITHIN A STRUCTURED BUSINESS ORGANIZATION UNDER TIME, RESOURCE AND BUDGET CONSTRAINTS
Issued US PCT/US2015/022448
Aspects of the present disclosure are presented for efficiently allocating resources to projects in a schedule under time, resource and budget constraints. In some embodiments, a method is presented. The method may include accessing variables for determining an efficient allocation of resources in the schedule, including a set of project dependency values indicating which projects in the plurality of projects must be completed as requisite for completing other projects in the plurality of…
Aspects of the present disclosure are presented for efficiently allocating resources to projects in a schedule under time, resource and budget constraints. In some embodiments, a method is presented. The method may include accessing variables for determining an efficient allocation of resources in the schedule, including a set of project dependency values indicating which projects in the plurality of projects must be completed as requisite for completing other projects in the plurality of projects. The method may also include determining a dependency path indicating an ordering of projects to be completed, based on the set of project dependency values, wherein a project in the dependency path cannot be started until all preceding projects in the dependency path are completed; and determining an efficient selection of projects to be completed within the time horizon that maximizes an optimization goal, based on the dependency path and constrained by budget expenditures.
Other inventors -
Projects
-
GigWme projects
-
Understand how interest netwroks maps over time and location.
Honors & Awards
-
Fulbright fellowship
Fulbright Organization
-
Parallel Computing Fellowship
NATO
Languages
-
Italian
-
Organizations
-
INFORMS
-
- Present
Recommendations received
2 people have recommended Cosimo
Join now to viewOther similar profiles
Explore more posts
Explore top content on LinkedIn
Find curated posts and insights for relevant topics all in one place.
View top content