We introduce Granite Vision, a lightweight large language model with vision capabilities, specifically designed to excel in enterprise use cases, particularly in visual document understanding. Our model is trained on a comprehensive instruction-following dataset, including document-related tasks, such as content extraction from tables, charts, diagrams, sketches, and infographics, as well as general image tasks. The architecture of Granite Vision is centered around visual modality alignment with a decoder-only, 2 billion parameter Granite large language model. Additionally, we introduce a dedicated safety classification approach in test-time that leverages a sparse set of attention vectors to identify potential harmful inputs. Despite its lightweight architecture, Granite Vision achieves strong results in standard benchmarks related to visual document understanding, as well as on the LiveXiv benchmark, which is designed to avoid test set contamination by using a constantly updated corpus of recently published Arxiv papers. We are releasing the model under the Apache-2 license, allowing for both research and commercial use, while offering complete visibility into the training data and other relevant details. See https://huggingface.co/ibm-granite/ for model weights.

Artificial intelligence (AI) is defined as the ability of machines to perform tasks that are usually associated with intelligent beings. Argument and debate are fundamental capabilities of human intelligence, essential for a wide range of human activities, and common to all human societies. The development of computational argumentation technologies is therefore an important emerging discipline in AI research 1 . Here we present Project Debater, an autonomous debating system that can engage in a competitive debate with humans. We provide a complete description of the system’s architecture, a thorough and systematic evaluation of its operation across a wide range of debate topics, and a detailed account of the system’s performance in its public debut against three expert human debaters. We also highlight the fundamental differences between debating with humans as opposed to challenging humans in game competitions, the latter being the focus of classical ‘grand challenges’ pursued by the AI research community over the past few decades. We suggest that such challenges lie in the ‘comfort zone’ of AI, whereas debating with humans lies in a different territory, in which humans still prevail, and for which novel paradigms are required to make substantial progress.

In February 2011, the world was introduced to Watson, IBM’s cognitive computing system [1] that defeated Ken Jennings and Brad Rutter at Jeopardy! It was the first widely seen demonstration of cognitive computing, and it marked the end of the so-called“AI winter.”Watson’s ability to answer subtle, complex, pun-laden questions made clear that a new era of computing was at hand. An era where computers can start making sense of the vast amount of unstructured data in the world and apply this understanding not only to answer trivia questions, but also to tackle some of the world’s pressing problems and change how people interact with computers. Indeed, since Jeopardy!, Watson has tackled increasingly complex data sets, and developed understanding, reasoning, and learning. Specifically, we have identified five core capabilities of Cognitive Computing: 1. They create deeper human engagement; 2. They scale and elevate expertise; 3. They infuse products and services with cognition; 4. They enable cognitive processes and operations; 5. They enhance exploration and discovery. The true potential of the Cognitive Era will be realized by combining the data analytics, and statistical reasoning of machines with uniquely human qualities, such as self-directed goals, common sense, and ethical values. This is what Watson was built to do, and is in fact already doing. Banks are analyzing customer requests and financial data to surface insights to help them make investment recommendations. Companies in heavily regulated industries are querying the system to keep up with ever-changing legislation and standards of compliance. And oncologists are testing ways in which cognitive systems can help interpret cancer patients’ clinical information and identify individualized, evidence-based treatment options that leverage specialists’ experience and research. Customer and technical support are in particular being re-invented, relying more on cognitive systems for self-help and for agent assist. In this talk, we highlight some of the applications of Watson that are being pursued by IBM. These include applications in customer and technical support, in Finance, in Legal

Measuring the effectiveness of marketing campaigns across different channels is one of the most challenging tasks for today's brand marketers. Such measurement usually relies on a combination of key performance indicators (KPIs), used for assessing various aspects of marketing outcomes. Recently, with the availability of social-media sources, new options for collecting KPIs have become available and numerous social-media monitoring tools and services have emerged. Yet, given the vast media spectrum, which goes beyond social-media channels, existing solutions fail to generalize well and the curation of marketing performance KPIs for most marketing channels still relies on labor intensive means such as surveys and questionnaires. Trying to address the challenges, we propose to demonstrate a novel solution we have developed in IBM: Multi-channel Marketing Monitoring Platform (M3P for short). M3P is better tailored for the marketing performance domain, where online chatter is being harnessed for effective collection of meaningful marketing KPIs across all possible channels. We describe M3P's main challenges and review some of its novel KPIs. We then describe the M3P solution, focusing on its KPI extraction process. Finally, we describe the planned demonstration using a real-world marketing use-case.

Information integration is the cornerstone of modern business informatics. It is a pervasive problem; rarely is a new application built without an initial phase of gathering and integrating information. Information integration comes in a wide variety of forms. Historically, two major approaches were recognized: data federation and data warehousing. Today, we need new approaches, as information integration becomes more dynamic, while coping with growing volumes of increasingly dirty and diverse data. At the same time, information integration must be coupled more tightly with the applications and the analytics that will leverage the integrated results, to make the integration process more tractable and the results more consumable.

This research explores new ways to augment the search and discovery of relations between Web 2.0 entities using multiple types and sources of social information. Our goal is to allow the search for all object types such as documents, persons and tags, while retrieving related objects of all types. We implemented a social-search engine using a unified approach, where the search space is expanded to represent heterogeneous information objects that are interrelated by several relation types. Our solution is based on multifaceted search, which provides an efficient update mechanism for relations between objects, as well as efficient search over the heterogeneous data. We describe a social search engine positioned within a large enterprise, applied over social data gathered from several Web 2.0 applications. We conducted a large user study with over 600 people to evaluate the contribution of social data for search. Our results demonstrate the high precision of social search results and confirm the strong relationship of users and tags to the topics retrieved.

We describe a system for automating call-center analysis and monitoring. Our system integrates transcription of incoming calls with analysis of their content; for the analysis, we introduce a novel method of estimating the domain-specific importance of conversation fragments, based on divergence of corpus statistics. Combining this method with Information Retrieval approaches, we provide knowledge-mining tools both for the call-center agents and for administrators of the center.

Biomedical text plays a fundamental role in knowledge discovery in life science, in both basic research (in the field of bioinformatics) and in industry sectors devoted to improving medical practice, drug development, and health care (such as medical informatics, clinical genomics, and other sectors). Several groups in the IBM Research Division are collaborating on the development of a prototype system for text analysis, search, and text-mining methods to support problem solving in life science. The system is called "BioTeKS" ("Biological Text Knowledge Services"), and it integrates research technologies from multiple IBM Research labs. BioTeKS is also the first major application of the UIMA (Unstructured Information Management Architecture) initiative also emerging from IBM Research. BioTeKS is intended to analyze biomedical text such as MEDLINETM abstracts, medical records, and patents; text is analyzed by automatically identifying terms or names corresponding to key biomedical entities (e.g., "genes," "proteins," "compounds," or "drugs") and concepts or facts related to them. In this paper, we describe the value of text analysis in biomedical research, the development of the BioTeKS system, and applications which demonstrate its functions.

Although time has been recognized as an important dimension in the co-citation literature, to date it has not been incorporated into the analogous process of link analysis on the Web. In this paper, we discuss several aspects and uses of the time dimension in the context of Web information retrieval. We describe the ideal case where search engines track and store temporal data for each of the pages in their repository, assigning timestamps to the hyperlinks embedded within the pages. We introduce several applications which benefit from the availability of such timestamps. To demonstrate our claims, we use a somewhat simplistic approach, which dates links by approximating the age of the page's content. We show that by using this crude measure alone it is possible to detect and expose significant events and trends. We predict that by using more robust methods for tracking modifications in the content of pages, search engines will be able to provide results that are more timely and better reflect current real-life trends than those they provide today.

We describe Web-a-Where, a system for associating geography with Web pages. Web-a-Where locates mentions of places and determines the place each name refers to. In addition, it assigns to each page a geographic focus --- a locality that the page discusses as a whole. The tagging process is simple and fast, aimed to be applied to large collections of Web pages and to facilitate a variety of location-based applications and data analyses.Geotagging involves arbitrating two types of ambiguities: geo/non-geo and geo/geo. A geo/non-geo ambiguity occurs when a place name also has a non-geographic meaning, such as a person name (e.g., Berlin) or a common word (Turkey). Geo/geo ambiguity arises when distinct places have the same name, as in London, England vs. London, Ontario.An implementation of the tagger within the framework of the WebFountain data mining system is described, and evaluated on several corpora of real Web pages. Precision of up to 82% on individual geotags is achieved. We also evaluate the relative contribution of various heuristics the tagger employs, and evaluate the focus-finding algorithm using a corpus pretagged with localities, showing that as many as 91% of the foci reported are correct up to the country level.

This paper describes an evaluation method based on Term Relevance Sets Trels that measures an IR system's quality by examining the content of the retrieved results rather than by looking for pre-specified relevant pages. Trels consist of a list of terms believed to be relevant for a particular query as well as a list of irrelevant terms. The proposed method does not involve any document relevance judgments, and as such is not adversely affected by changes to the underlying collection. Therefore, it can better scale to very large, dynamic collections such as the Web. Moreover, this method can evaluate a system's effectiveness on an updatable "live" collection, or on collections derived from different data sources. Our experiments show that the proposed method is very highly correlated with official TREC measures.