Radu Marinescu

Daniel Karl I. Weidele, Shazia Afzal, Abel N. Valente et al.

We present AutoDOViz, an interactive user interface for automated decision optimization (AutoDO) using reinforcement learning (RL). Decision optimization (DO) has classically being practiced by dedicated DO researchers where experts need to spend long periods of time fine tuning a solution through trial-and-error. AutoML pipeline search has sought to make it easier for a data scientist to find the best machine learning pipeline by leveraging automation to search and tune the solution. More recently, these advances have been applied to the domain of AutoDO, with a similar goal to find the best reinforcement learning pipeline through algorithm selection and parameter tuning. However, Decision Optimization requires significantly more complex problem specification when compared to an ML problem. AutoDOViz seeks to lower the barrier of entry for data scientists in problem specification for reinforcement learning problems, leverage the benefits of AutoDO algorithms for RL pipeline search and finally, create visualizations and policy insights in order to facilitate the typical interactive nature when communicating problem formulation and solution proposals between DO experts and domain experts. In this paper, we report our findings from semi-structured expert interviews with DO practitioners as well as business consultants, leading to design requirements for human-centered automation for DO with RL. We evaluate a system implementation with data scientists and find that they are significantly more open to engage in DO after using our proposed solution. AutoDOViz further increases trust in RL agent models and makes the automated training and evaluation process more comprehensible. As shown for other automation in ML tasks, we also conclude automation of RL for DO can benefit from user and vice-versa when the interface promotes human-in-the-loop.

Jasmina Gajcin, James McCarthy, Rahul Nair et al.

A well-defined reward function is crucial for successful training of an reinforcement learning (RL) agent. However, defining a suitable reward function is a notoriously challenging task, especially in complex, multi-objective environments. Developers often have to resort to starting with an initial, potentially misspecified reward function, and iteratively adjusting its parameters, based on observed learned behavior. In this work, we aim to automate this process by proposing ITERS, an iterative reward shaping approach using human feedback for mitigating the effects of a misspecified reward function. Our approach allows the user to provide trajectory-level feedback on agent's behavior during training, which can be integrated as a reward shaping signal in the following training iteration. We also allow the user to provide explanations of their feedback, which are used to augment the feedback and reduce user effort and feedback frequency. We evaluate ITERS in three environments and show that it can successfully correct misspecified reward functions.

Bobak Pezeshki, Radu Marinescu, Alexander Ihler et al.

Scientific computing has experienced a surge empowered by advancements in technologies such as neural networks. However, certain important tasks are less amenable to these technologies, benefiting from innovations to traditional inference schemes. One such task is protein re-design. Recently a new re-design algorithm, AOBB-K*, was introduced and was competitive with state-of-the-art BBK* on small protein re-design problems. However, AOBB-K* did not scale well. In this work we focus on scaling up AOBB-K* and introduce three new versions: AOBB-K*-b (boosted), AOBB-K*-DH (with dynamic heuristics), and AOBB-K*-UFO (with underflow optimization) that significantly enhance scalability.

James McCarthy, Rahul Nair, Elizabeth Daly et al.

Explainability of Reinforcement Learning (RL) policies remains a challenging research problem, particularly when considering RL in a safety context. Understanding the decisions and intentions of an RL policy offer avenues to incorporate safety into the policy by limiting undesirable actions. We propose the use of a Boolean Decision Rules model to create a post-hoc rule-based summary of an agent's policy. We evaluate our proposed approach using a DQN agent trained on an implementation of a lava gridworld and show that, given a hand-crafted feature representation of this gridworld, simple generalised rules can be created, giving a post-hoc explainable summary of the agent's policy. We discuss possible avenues to introduce safety into a RL agent's policy by using rules generated by this rule-based model as constraints imposed on the agent's policy, as well as discuss how creating simple rule summaries of an agent's policy may help in the debugging process of RL agents.

Javier Carnerero-Cano, Massimiliano Pronesti, Radu Marinescu et al.

Large language models (LLMs) are widely used in knowledge-intensive applications but often generate factually incorrect responses. A promising approach to rectify these flaws is correcting LLMs using feedback. Therefore, in this paper, we introduce FactCorrector, a new post-hoc correction method that adapts across domains without retraining and leverages structured feedback about the factuality of the original response to generate a correction. To support rigorous evaluations of factuality correction methods, we also develop the VELI5 benchmark, a novel dataset containing systematically injected factual errors and ground-truth corrections. Experiments on VELI5 and several popular long-form factuality datasets show that the FactCorrector approach significantly improves factual precision while preserving relevance, outperforming strong baselines. We release our code at https://ibm.biz/factcorrector.

Radu Marinescu, Debarun Bhattacharjya, Junkyu Lee et al.

Large language models (LLMs) have achieved remarkable success in generative tasks, yet they often fall short in ensuring the factual accuracy of their outputs, thus limiting their reliability in real-world applications where correctness is critical. In this paper, we present FactReasoner, a novel neuro-symbolic based factuality assessment framework that employs probabilistic reasoning to evaluate the truthfulness of long-form generated responses. FactReasoner decomposes a response into atomic units, retrieves relevant contextual information from external knowledge sources, and models the logical relationships (e.g., entailment, contradiction) between these units and their contexts using probabilistic encodings. It then estimates the posterior probability that each atomic unit is supported by the retrieved evidence. Our experiments on both labeled and unlabeled benchmark datasets demonstrate that FactReasoner often outperforms state-of-the-art prompt-based methods in terms of factual precision and recall. Our open-source implementation is publicly available at: https://github.com/IBM/FactReasoner.

Yufang Hou, Alessandra Pascale, Javier Carnerero-Cano et al.

Retrieval-augmented generation (RAG) has emerged as a promising solution to mitigate the limitations of large language models (LLMs), such as hallucinations and outdated information. However, it remains unclear how LLMs handle knowledge conflicts arising from different augmented retrieved passages, especially when these passages originate from the same source and have equal trustworthiness. In this work, we conduct a comprehensive evaluation of LLM-generated answers to questions that have varying answers based on contradictory passages from Wikipedia, a dataset widely regarded as a high-quality pre-training resource for most LLMs. Specifically, we introduce WikiContradict, a benchmark consisting of 253 high-quality, human-annotated instances designed to assess LLM performance when augmented with retrieved passages containing real-world knowledge conflicts. We benchmark a diverse range of both closed and open-source LLMs under different QA scenarios, including RAG with a single passage, and RAG with 2 contradictory passages. Through rigorous human evaluations on a subset of WikiContradict instances involving 5 LLMs and over 3,500 judgements, we shed light on the behaviour and limitations of these models. For instance, when provided with two passages containing contradictory facts, all models struggle to generate answers that accurately reflect the conflicting nature of the context, especially for implicit conflicts requiring reasoning. Since human evaluation is costly, we also introduce an automated model that estimates LLM performance using a strong open-source language model, achieving an F-score of 0.8. Using this automated metric, we evaluate more than 1,500 answers from seven LLMs across all WikiContradict instances. To facilitate future work, we release WikiContradict on: https://ibm.biz/wikicontradict.

Debarun Bhattacharjya, Junkyu Lee, Don Joven Agravante et al.

Foundation models (FMs) such as large language models have revolutionized the field of AI by showing remarkable performance in various tasks. However, they exhibit numerous limitations that prevent their broader adoption in many real-world systems, which often require a higher bar for trustworthiness and usability. Since FMs are trained using loss functions aimed at reconstructing the training corpus in a self-supervised manner, there is no guarantee that the model's output aligns with users' preferences for a specific task at hand. In this survey paper, we propose a conceptual framework that encapsulates different modes by which agents could interact with FMs and guide them suitably for a set of tasks, particularly through knowledge augmentation and reasoning. Our framework elucidates agent role categories such as updating the underlying FM, assisting with prompting the FM, and evaluating the FM output. We also categorize several state-of-the-art approaches into agent interaction protocols, highlighting the nature and extent of involvement of the various agent roles. The proposed framework provides guidance for future directions to further realize the power of FMs in practical AI systems.

Jasmina Gajcin, Erik Miehling, Rahul Nair et al.

Using LLMs to evaluate text, that is, LLM-as-a-judge, is increasingly being used at scale to augment or even replace human annotations. As such, it is imperative that we understand the potential biases and risks of doing so. In this work, we propose an approach for extracting high-level concept-based global policies from LLM-as-a-Judge. Our approach consists of two algorithms: 1) CLoVE (Contrastive Local Verifiable Explanations), which generates verifiable, concept-based, contrastive local explanations and 2) GloVE (Global Verifiable Explanations), which uses iterative clustering, summarization and verification to condense local rules into a global policy. We evaluate GloVE on seven standard benchmarking datasets for content harm detection. We find that the extracted global policies are highly faithful to decisions of the LLM-as-a-Judge. Additionally, we evaluated the robustness of global policies to text perturbations and adversarial attacks. Finally, we conducted a user study to evaluate user understanding and satisfaction with global policies.

Quan Xiao, Debarun Bhattacharjya, Balaji Ganesan et al.

Estimating the confidence of large language model (LLM) outputs is essential for real-world applications requiring high user trust. Black-box uncertainty quantification (UQ) methods, relying solely on model API access, have gained popularity due to their practical benefits. In this paper, we examine the implicit assumption behind several UQ methods, which use generation consistency as a proxy for confidence, an idea we formalize as the consistency hypothesis. We introduce three mathematical statements with corresponding statistical tests to capture variations of this hypothesis and metrics to evaluate LLM output conformity across tasks. Our empirical investigation, spanning 8 benchmark datasets and 3 tasks (question answering, text summarization, and text-to-SQL), highlights the prevalence of the hypothesis under different settings. Among the statements, we highlight the `Sim-Any' hypothesis as the most actionable, and demonstrate how it can be leveraged by proposing data-free black-box UQ methods that aggregate similarities between generations for confidence estimation. These approaches can outperform the closest baselines, showcasing the practical value of the empirically observed consistency hypothesis.

Debarun Bhattacharjya, Balaji Ganesan, Junkyu Lee et al.

When does a large language model (LLM) know what it does not know? Uncertainty quantification (UQ) provides measures of uncertainty, such as an estimate of the confidence in an LLM's generated output, and is therefore increasingly recognized as a crucial component of trusted AI systems. Black-box UQ methods do not require access to internal model information from the generating LLM and therefore have numerous real-world advantages, such as robustness to system changes, adaptability to choice of LLM, reduced costs, and computational tractability. In this paper, we investigate the effectiveness of UQ techniques that are primarily but not necessarily entirely black-box, where the consistency between a generated output and other sampled generations is used as a proxy for confidence in its correctness. We propose a high-level non-verbalized similarity-based aggregation framework that subsumes a broad swath of UQ approaches suitable for complex generative tasks, as well as introduce specific novel techniques from the framework that train confidence estimation models using small training sets. Through an empirical study with datasets spanning the diverse tasks of question answering, summarization, and text-to-SQL, we demonstrate that our proposed similarity-based methods can yield better calibrated confidences than baselines.

João P. Arroyo, João G. Rodrigues, Daniel Lawand et al.

We investigate partially identifiable queries in a class of causal models. We focus on acyclic Structural Causal Models that are quasi-Markovian (that is, each endogenous variable is connected with at most one exogenous confounder). We look into scenarios where endogenous variables are observed (and a distribution over them is known), while exogenous variables are not fully specified. This leads to a representation that is in essence a Bayesian network where the distribution of root variables is not uniquely determined. In such circumstances, it may not be possible to precisely compute a probability value of interest. We thus study the computation of tight probability bounds, a problem that has been solved by multilinear programming in general, and by linear programming when a single confounded component is intervened upon. We present a new algorithm to simplify the construction of such programs by exploiting input probabilities over endogenous variables. For scenarios with a single intervention, we apply column generation to compute a probability bound through a sequence of auxiliary linear integer programs, thus showing that a representation with polynomial cardinality for exogenous variables is possible. Experiments show column generation techniques to be superior to existing methods.

James McCarthy, Radu Marinescu, Elizabeth Daly et al.

Risk-averse Constrained Reinforcement Learning (RaCRL) aims to learn policies that minimise the likelihood of rare and catastrophic constraint violations caused by an environment's inherent randomness. In general, risk-aversion leads to conservative exploration of the environment which typically results in converging to sub-optimal policies that fail to adequately maximise reward or, in some cases, fail to achieve the goal. In this paper, we propose an exploration-based approach for RaCRL called Optimistic Risk-averse Actor Critic (ORAC), which constructs an exploratory policy by maximising a local upper confidence bound of the state-action reward value function whilst minimising a local lower confidence bound of the risk-averse state-action cost value function. Specifically, at each step, the weighting assigned to the cost value is increased or decreased if it exceeds or falls below the safety constraint value. This way the policy is encouraged to explore uncertain regions of the environment to discover high reward states whilst still satisfying the safety constraints. Our experimental results demonstrate that the ORAC approach prevents convergence to sub-optimal policies and improves significantly the reward-cost trade-off in various continuous control tasks such as Safety-Gymnasium and a complex building energy management environment CityLearn.

Achille Fokoue, Ibrahim Abdelaziz, Maxwell Crouse et al.

Using reinforcement learning for automated theorem proving has recently received much attention. Current approaches use representations of logical statements that often rely on the names used in these statements and, as a result, the models are generally not transferable from one domain to another. The size of these representations and whether to include the whole theory or part of it are other important decisions that affect the performance of these approaches as well as their runtime efficiency. In this paper, we present NIAGRA; an ensemble Name InvAriant Graph RepresentAtion. NIAGRA addresses this problem by using 1) improved Graph Neural Networks for learning name-invariant formula representations that is tailored for their unique characteristics and 2) an efficient ensemble approach for automated theorem proving. Our experimental evaluation shows state-of-the-art performance on multiple datasets from different domains with improvements up to 10% compared to the best learning-based approaches. Furthermore, transfer learning experiments show that our approach significantly outperforms other learning-based approaches by up to 28%.

Jasmina Gajcin, Rahul Nair, Tejaswini Pedapati et al.

In complex tasks where the reward function is not straightforward and consists of a set of objectives, multiple reinforcement learning (RL) policies that perform task adequately, but employ different strategies can be trained by adjusting the impact of individual objectives on reward function. Understanding the differences in strategies between policies is necessary to enable users to choose between offered policies, and can help developers understand different behaviors that emerge from various reward functions and training hyperparameters in RL systems. In this work we compare behavior of two policies trained on the same task, but with different preferences in objectives. We propose a method for distinguishing between differences in behavior that stem from different abilities from those that are a consequence of opposing preferences of two RL agents. Furthermore, we use only data on preference-based differences in order to generate contrasting explanations about agents' preferences. Finally, we test and evaluate our approach on an autonomous driving task and compare the behavior of a safety-oriented policy and one that prefers speed.

Haifeng Qian, Radu Marinescu, Alexander Gray et al.

This paper introduces Logical Credal Networks, an expressive probabilistic logic that generalizes many prior models that combine logic and probability. Given imprecise information represented by probability bounds and conditional probability bounds of logic formulas, this logic specifies a set of probability distributions over all interpretations. On the one hand, our approach allows propositional and first-order logic formulas with few restrictions, e.g., without requiring acyclicity. On the other hand, it has a Markov condition similar to Bayesian networks and Markov random fields that is critical in real-world applications. Having both these properties makes this logic unique, and we investigate its performance on maximum a posteriori inference tasks, including solving Mastermind games with uncertainty and detecting credit card fraud. The results show that the proposed method outperforms existing approaches, and its advantage lies in aggregating multiple sources of imprecise information.

Paulito P. Palmes, Akihiro Kishimoto, Radu Marinescu et al.

The pipeline optimization problem in machine learning requires simultaneous optimization of pipeline structures and parameter adaptation of their elements. Having an elegant way to express these structures can help lessen the complexity in the management and analysis of their performances together with the different choices of optimization strategies. With these issues in mind, we created the AutoMLPipeline (AMLP) toolkit which facilitates the creation and evaluation of complex machine learning pipeline structures using simple expressions. We use AMLP to find optimal pipeline signatures, datamine them, and use these datamined features to speed-up learning and prediction. We formulated a two-stage pipeline optimization with surrogate modeling in AMLP which outperforms other AutoML approaches with a 4-hour time budget in less than 5 minutes of AMLP computation time.

Adi Botea, Christian Muise, Shubham Agarwal et al.

Dialogue systems have many applications such as customer support or question answering. Typically they have been limited to shallow single turn interactions. However more advanced applications such as career coaching or planning a trip require a much more complex multi-turn dialogue. Current limitations of conversational systems have made it difficult to support applications that require personalization, customization and context dependent interactions. We tackle this challenging problem by using domain-independent AI planning to automatically create dialogue plans, customized to guide a dialogue towards achieving a given goal. The input includes a library of atomic dialogue actions, an initial state of the dialogue, and a goal. Dialogue plans are plugged into a dialogue system capable to orchestrate their execution. Use cases demonstrate the viability of the approach. Our work on dialogue planning has been integrated into a product, and it is in the process of being deployed into another.

Robert Mateescu, Rina Dechter, Radu Marinescu

Inspired by the recently introduced framework of AND/OR search spaces for graphical models, we propose to augment Multi-Valued Decision Diagrams (MDD) with AND nodes, in order to capture function decomposition structure and to extend these compiled data structures to general weighted graphical models (e.g., probabilistic models). We present the AND/OR Multi-Valued Decision Diagram (AOMDD) which compiles a graphical model into a canonical form that supports polynomial (e.g., solution counting, belief updating) or constant time (e.g. equivalence of graphical models) queries. We provide two algorithms for compiling the AOMDD of a graphical model. The first is search-based, and works by applying reduction rules to the trace of the memory intensive AND/OR search algorithm. The second is inference-based and uses a Bucket Elimination schedule to combine the AOMDDs of the input functions via the the APPLY operator. For both algorithms, the compilation time and the size of the AOMDD are, in the worst case, exponential in the treewidth of the graphical model, rather than pathwidth as is known for ordered binary decision diagrams (OBDDs). We introduce the concept of semantic treewidth, which helps explain why the size of a decision diagram is often much smaller than the worst case bound. We provide an experimental evaluation that demonstrates the potential of AOMDDs.

Radu Marinescu, Kalev Kask, Rina Dechter

The paper continues the study of partitioning based inference of heuristics for search in the context of solving the Most Probable Explanation task in Bayesian Networks. We compare two systematic Branch and Bound search algorithms, BBBT (for which the heuristic information is constructed during search and allows dynamic variable/value ordering) and its predecessor BBMB (for which the heuristic information is pre-compiled), against a number of popular local search algorithms for the MPE problem. We show empirically that, when viewed as approximation schemes, BBBT/BBMB are superior to all of these best known SLS algorithms, especially when the domain sizes increase beyond 2. This is in contrast with the performance of SLS vs. systematic search on CSP/SAT problems, where SLS often significantly outperforms systematic algorithms. As far as we know, BBBT/BBMB are currently the best performing algorithms for solving the MPE task.

Radu Marinescu, Abdul Razak, Nic Wilson

We describe multi-objective influence diagrams, based on a set of p objectives, where utility values are vectors in Rp, and are typically only partially ordered. These can still be solved by a variable elimination algorithm, leading to a set of maximal values of expected utility. If the Pareto ordering is used this set can often be prohibitively large. We consider approximate representations of the Pareto set based on e-coverings, allowing much larger problems to be solved. In addition, we define a method for incorporating user tradeoffs, which also greatly improves the efficiency.

Radu Marinescu, Rina Dechter

The paper evaluates the power of best-first search over AND/OR search spaces for solving the Most Probable Explanation (MPE) task in Bayesian networks. The main virtue of the AND/OR representation of the search space is its sensitivity to the structure of the problem, which can translate into significant time savings. In recent years depth-first AND/OR Branch-and- Bound algorithms were shown to be very effective when exploring such search spaces, especially when using caching. Since best-first strategies are known to be superior to depth-first when memory is utilized, exploring the best-first control strategy is called for. The main contribution of this paper is in showing that a recent extension of AND/OR search algorithms from depth-first Branch-and-Bound to best-first is indeed very effective for computing the MPE in Bayesian networks. We demonstrate empirically the superiority of the best-first search approach on various probabilistic networks.

Radu Marinescu, Nic Wilson

In this paper, we develop a qualitative theory of influence diagrams that can be used to model and solve sequential decision making tasks when only qualitative (or imprecise) information is available. Our approach is based on an order-of-magnitude approximation of both probabilities and utilities and allows for specifying partially ordered preferences via sets of utility values. We also propose a dedicated variable elimination algorithm that can be applied for solving order-of-magnitude influence diagrams.