Organizations handling sensitive documents face a critical dilemma: adopt cloud-based AI systems that offer powerful question-answering capabilities but compromise data privacy, or maintain local processing that ensures security but delivers poor accuracy. We present a question-answering system that resolves this trade-off by combining semantic understanding with keyword precision, operating entirely on local infrastructure without internet access. Our approach demonstrates that organizations can achieve competitive accuracy on complex queries across legal, scientific, and conversational documents while keeping all data on their machines. By balancing two complementary retrieval strategies and using consumer-grade hardware acceleration, the system delivers reliable answers with minimal errors, letting banks, hospitals, and law firms adopt conversational document AI without transmitting proprietary information to external providers. This work establishes that privacy and performance need not be mutually exclusive in enterprise AI deployment.

Prior works in multi-objective reinforcement learning typically use linear reward scalarization with fixed weights, which provably fail to capture non-convex Pareto fronts and thus yield suboptimal results. This limitation becomes especially critical in online preference alignment for large language models. Here, stochastic trajectories generated by parameterized policies create highly non-linear and non-convex mappings from parameters to objectives that no single static weighting scheme can find optimal trade-offs. We address this limitation by introducing dynamic reward weighting, which adap-tively adjusts reward weights during the online reinforcement learning process. Unlike existing approaches that rely on fixed-weight interpolation, our dynamic weighting continuously balances and prioritizes objectives in training, facilitating effective exploration of Pareto fronts in objective space. We introduce two approaches of increasing sophistication and generalizability: (1) hypervolume-guided weight adaptation and (2) gradient-based weight optimization, offering a versatile toolkit for online multi-objective alignment. Our extensive experiments demonstrate their compatibility with commonly used online reinforcement learning algorithms (including GRPO, REINFORCE, and RLOO), effectiveness across multiple mathematical reasoning datasets, and applicability to different model families, consistently achieving Pareto dominant solutions with fewer training steps than fixed-weight linear scalarization baselines.Figure 1: Pareto fronts obtained by our gradient-based weight optimization compared to three baselines using fixed-weight reward interpolation.

In this work, we demonstrate that a simple two-layer neural network with standard activation functions can learn an arbitrary word operation in any finite group, provided sufficient width is available and exhibits grokking while doing so. To explain the mechanism by which this is achieved, we reframe the problem as that of learning a particular $3$-tensor, which we show is typically of low rank. A key insight is that low-rank implementations of this tensor can be obtained by decomposing it along triplets of basic self-conjugate representations of the group and leveraging the fusion structure to rule out many components. Focusing on a phenomenologically similar but more tractable surrogate model, we show that the network is able to find such low-rank implementations (or approximations thereof), thereby using limited width to approximate the word-tensor in a generalizable way. In the case of the simple multiplication word, we further elucidate the form of these low-rank implementations, showing that the network effectively implements efficient matrix multiplication in the sense of Strassen. Our work also sheds light on the mechanism by which a network reaches such a solution under gradient descent.

Structure-based drug design (SBDD) leverages the 3D structure of biomolecular targets to guide the creation of new therapeutic agents. Recent advances in generative models, including diffusion models and geometric deep learning, have demonstrated promise in optimizing ligand generation. However, the scarcity of high-quality protein-ligand complex data and the inherent challenges in aligning generated ligands with target proteins limit the effectiveness of these methods. We propose BoKDiff, a novel framework that enhances ligand generation by combining multi-objective optimization and Best-of-K alignment methodologies. Built upon the DecompDiff model, BoKDiff generates diverse candidates and ranks them using a weighted evaluation of molecular properties such as QED, SA, and docking scores. To address alignment challenges, we introduce a method that relocates the center of mass of generated ligands to their docking poses, enabling accurate sub-component extraction. Additionally, we integrate a Best-of-N (BoN) sampling approach, which selects the optimal ligand from multiple generated candidates without requiring fine-tuning. BoN achieves exceptional results, with QED values exceeding 0.6, SA scores above 0.75, and a success rate surpassing 35%, demonstrating its efficiency and practicality. BoKDiff achieves state-of-the-art results on the CrossDocked2020 dataset, including a -8.58 average Vina docking score and a 26% success rate in molecule generation. This study is the first to apply Best-of-K alignment and Best-of-N sampling to SBDD, highlighting their potential to bridge generative modeling with practical drug discovery requirements. The code is provided at https://github.com/khodabandeh-ali/BoKDiff.git.

The following content is mainly about the article Avoiding Catastrophe: Active Dendrites Enable Multi-Tasking Learning in Dynamic Environments by A. Iyer et al. (December 2021). It is a pleasant paper mixing biology, neuroscience and mathematical modeling, I hope you find it interesting. Standard Artificial Neural Networks (ANNs), based on the (inaccurate) point neuron model [Lapique, 1907] and backpropagation algorithm, often fail dramatically in multiple task learning. Differently from single-task machine learning, learning multiple distinct tasks introduces new complications. When using gradient-based methods (such as backpropagation), a noteworthy issue is that error gradients and accumulated knowledge from different tasks can interfere with one another.

Geometric structure of an optimization landscape is argued to be fundamentally important to support the success of deep learning. However, recent research efforts focused on either of toy random models with unrealistic assumptions and numerical evidences about different shapes of the optimization landscape, thereby lacking a unified view about the nature of the landscape. Here, we propose a statistical mechanics framework by directly building a least structured model of the high-dimensional weight space, considering realistic structured data, stochastic gradient descent algorithms, and the computational depth of the network parametrized by weight parameters. We also consider whether the number of network parameters outnumbers the number of supplied training data, namely, over- or under-parametrization. Our least structured model predicts that the weight spaces of the under-parametrization and over-parameterization cases belong to the same class. These weight spaces are well-connected without any heterogeneous geometric properties. In contrast, the shallow-network has a shattered weight space, characterized by discontinuous phase transitions in physics, thereby clarifying roles of depth in deep learning. Our effective model also predicts that inside a deep network, there exists a liquid-like central part of the architecture in the sense that the weights in this part behave as randomly as possible. Our work may thus explain why deep learning is unreasonably effective in terms of the high-dimensional weight space, and how deep networks are different from shallow ones.

This is the second part of the neural network tutorial. The first part can be found here: https://link.medium.com/YCEAECVp0W Now that we have seen how a neural network is represented, we can go on to see how exactly it works. Since there are many layers having many neurons, there exists a complex set of weights to get an output from some input variables. Each weight in this network can be changed and hence there are countless configurations a neural network can have.

Reinforcement learning is widely used for dialogue policy optimization where the reward function often consists of more than one component, e.g., the dialogue success and the dialogue length. In this work, we propose a structured method for finding a good balance between these components by searching for the optimal reward component weighting. To render this search feasible, we use multi-objective reinforcement learning to significantly reduce the number of training dialogues required. We apply our proposed method to find optimized component weights for six domains and compare them to a default baseline.