Related papers: A Collision-Free Hot-Tier Extension for Engram-Sty…
The Engram module -- a hash-keyed, O(1) associative memory injected into Transformer layers -- was recently shown to improve large language model pretraining, with the appealing interpretation that it provides a content-addressed shortcut…
While Mixture-of-Experts (MoE) scales capacity via conditional computation, Transformers lack a native primitive for knowledge lookup, forcing them to inefficiently simulate retrieval through computation. To address this, we introduce…
Scaling conditional memory offers a promising way to increase language-model capacity, but existing methods such as Engram learn large memory tables from scratch during pre-training, making memory scaling expensive and sometimes…
Recent studies introduce conditional memory modules that decouple knowledge storage from neural computation, enabling more direct knowledge access. Compared to MoE, which relies on dynamic computation paths, explicit lookup provides a more…
Exemplar-Free Class Incremental Learning (EFCIL) aims to learn from a sequence of tasks without having access to previous task data. In this paper, we consider the challenging Cold Start scenario in which insufficient data is available in…
The safety alignment of large language models (LLMs) often relies on reinforcement learning from human feedback (RLHF), which requires human annotations to construct preference datasets. Given the challenge of assigning overall quality…
Exemplar-free Class Incremental Learning (EFCIL) aims to learn from a sequence of tasks without having access to previous task data. In this paper, we consider the challenging Cold Start scenario in which insufficient data is available in…
End-to-end agent-memory benchmarks report a single hit@k per retriever, confounding lexical leakage (uncontrolled query/gold/distractor entity overlap) with tag-mixing (preferences, services, tools averaged together). We propose…
Continual model merging integrates independently fine-tuned models sequentially without access to the original training data, offering a scalable and efficient solution for continual learning. However, existing methods face two critical…
The ability to learn continuously in artificial neural networks (ANNs) is often limited by catastrophic forgetting, a phenomenon in which new knowledge becomes dominant. By taking mechanisms of memory encoding in neuroscience (aka. engrams)…
Lane changes are complex safety and throughput critical driver actions. Most lane changing models deal with lane-changing maneuvers solely from the merging driver's standpoint and thus ignore driver interaction. To overcome this…
Stemming on the idea that a key objective in reinforcement learning is to invert a target distribution of effects, end-effect drives are proposed as an effective way to implement goal-directed motor learning, in the absence of an explicit…
Deep neural networks often rely on spurious features to make predictions, which makes them brittle under distribution shift and on samples where the spurious correlation does not hold (e.g., minority-group examples). Recent studies have…
A variety of statistical and machine learning methods are used to model crash frequency on specific roadways with machine learning methods generally having a higher prediction accuracy. Recently, heterogeneous ensemble methods (HEM),…
A key characteristic of deep recommendation models is the immense memory requirements of their embedding tables. These embedding tables can often reach hundreds of gigabytes which increases hardware requirements and training cost. A common…
Aligning generative recommender systems to user preferences via post-training is critical for closing the gap between next-item prediction and actual recommendation quality. Existing post-training methods are ill-suited for production-scale…
We develop a probabilistic framework for analysing model-based reinforcement learning in the episodic setting. We then apply it to study finite-time horizon stochastic control problems with linear dynamics but unknown coefficients and…
In this paper we discuss the problem of generically finding near-collisions for cryptographic hash functions in a memoryless way. A common approach is to truncate several output bits of the hash function and to look for collisions of this…
Long Short-Term Memory (LSTM) is one of the most widely used recurrent structures in sequence modeling. It aims to use gates to control information flow (e.g., whether to skip some information or not) in the recurrent computations, although…
Most Continual Learning (CL) methods maintain performance on earlier tasks by storing exemplars in a replay buffer, introducing memory overhead that scales with the number of tasks and raising privacy concerns in regulated domains. We…