The safety alignment of current Large Language Models (LLMs) is vulnerable. Simple attacks, or even benign fine-tuning, can jailbreak aligned models. We note that many of these vulnerabilities are related to a shared underlying issue: safety alignment can take shortcuts, wherein the alignment adapts a model's generative distribution primarily over only its very first few output tokens. We unifiedly refer to this issue as shallow safety alignment. In this paper, we present case studies to explain why shallow safety alignment can exist and show how this issue universally contributes to multiple recently discovered vulnerabilities in LLMs, including the susceptibility to adversarial suffix attacks, prefilling attacks, decoding parameter attacks, and fine-tuning attacks. The key contribution of this work is that we demonstrate how this consolidated notion of shallow safety alignment sheds light on promising research directions for mitigating these vulnerabilities. We show that deepening the safety alignment beyond the first few tokens can meaningfully improve robustness against some common exploits. We also design a regularized fine-tuning objective that makes the safety alignment more persistent against fine-tuning attacks by constraining updates on initial tokens. Overall, we advocate that future safety alignment should be made more than just a few tokens deep.

Prompting and in-context learning (ICL) have become efficient learning paradigms for large language models (LLMs). However, LLMs suffer from prompt brittleness and various bias factors in the prompt, including but not limited to the formatting, the choice verbalizers, and the ICL examples. To address this problem that results in unexpected performance degradation, calibration methods have been developed to mitigate the effects of these biases while recovering LLM performance. In this work, we first conduct a systematic analysis of the existing calibration methods, where we both provide a unified view and reveal the failure cases. Inspired by these analyses, we propose Batch Calibration (BC), a simple yet intuitive method that controls the contextual bias from the batched input, unifies various prior approaches and effectively addresses the aforementioned issues. BC is zero-shot, inference-only, and incurs negligible additional costs. In the few-shot setup, we further extend BC to allow it to learn the contextual bias from labeled data. We validate the effectiveness of BC with PaLM 2-(S, M, L) and CLIP models and demonstrate state-of-the-art performance over previous calibration baselines across more than 10 natural language understanding and image classification tasks.

Diagnosing and mitigating changes in model fairness under distribution shift is an important component of the safe deployment of machine learning in healthcare settings. Importantly, the success of any mitigation strategy strongly depends on the \textit{structure} of the shift. Despite this, there has been little discussion of how to empirically assess the structure of a distribution shift that one is encountering in practice. In this work, we adopt a causal framing to motivate conditional independence tests as a key tool for characterizing distribution shifts. Using our approach in two medical applications, we show that this knowledge can help diagnose failures of fairness transfer, including cases where real-world shifts are more complex than is often assumed in the literature. Based on these results, we discuss potential remedies at each step of the machine learning pipeline.

OBJECTIVE: Multitask learning (MTL) using electronic health records allows concurrent prediction of multiple endpoints. MTL has shown promise in improving model performance and training efficiency; however, it often suffers from negative transfer - impaired learning if tasks are not appropriately selected. We introduce a sequential subnetwork routing (SeqSNR) architecture that uses soft parameter sharing to find related tasks and encourage cross-learning between them. MATERIALS AND METHODS: Using the MIMIC-III (Medical Information Mart for Intensive Care-III) dataset, we train deep neural network models to predict the onset of 6 endpoints including specific organ dysfunctions and general clinical outcomes: acute kidney injury, continuous renal replacement therapy, mechanical ventilation, vasoactive medications, mortality, and length of stay. We compare single-task (ST) models with naive multitask and SeqSNR in terms of discriminative performance and label efficiency. RESULTS: SeqSNR showed a modest yet statistically significant performance boost across 4 of 6 tasks compared with ST and naive multitasking. When the size of the training dataset was reduced for a given task (label efficiency), SeqSNR outperformed ST for all cases showing an average area under the precision-recall curve boost of 2.1%, 2.9%, and 2.1% for tasks using 1%, 5%, and 10% of labels, respectively. CONCLUSIONS: The SeqSNR architecture shows superior label efficiency compared with ST and naive multitasking, suggesting utility in scenarios in which endpoint labels are difficult to ascertain.

We present a neuromorphic current mode implementation of a spiking neural classifier with lumped square law dendritic nonlinearity. It has been shown previously in software simulations that such a system with binary synapses can be trained with structural plasticity algorithms to achieve comparable classification accuracy with fewer synaptic resources than conventional algorithms. We show that even in real analog systems with manufacturing imperfections (CV of 23.5% and 14.4% for dendritic branch gains and leaks respectively), this network is able to produce comparable results with fewer synaptic resources. The chip fabricated in [Formula: see text]m complementary metal oxide semiconductor has eight dendrites per cell and uses two opposing cells per class to cancel common-mode inputs. The chip can operate down to a [Formula: see text] V and dissipates 19 nW of static power per neuronal cell and [Formula: see text] 125 pJ/spike. For two-class classification problems of high-dimensional rate encoded binary patterns, the hardware achieves comparable performance as software implementation of the same with only about a 0.5% reduction in accuracy. On two UCI data sets, the IC integrated circuit has classification accuracy comparable to standard machine learners like support vector machines and extreme learning machines while using two to five times binary synapses. We also show that the system can operate on mean rate encoded spike patterns, as well as short bursts of spikes. To the best of our knowledge, this is the first attempt in hardware to perform classification exploiting dendritic properties and binary synapses.

In this paper, we propose a novel winner-take-all (WTA) architecture employing neurons with nonlinear dendrites and an online unsupervised structural plasticity rule for training it. Furthermore, to aid hardware implementations, our network employs only binary synapses. The proposed learning rule is inspired by spike-timing-dependent plasticity but differs for each dendrite based on its activation level. It trains the WTA network through formation and elimination of connections between inputs and synapses. To demonstrate the performance of the proposed network and learning rule, we employ it to solve two-class, four-class, and six-class classification of random Poisson spike time inputs. The results indicate that by proper tuning of the inhibitory time constant of the WTA, a tradeoff between specificity and sensitivity of the network can be achieved. We use the inhibitory time constant to set the number of subpatterns per pattern we want to detect. We show that while the percentages of successful trials are 92%, 88%, and 82% for two-class, four-class, and six-class classification when no pattern subdivisions are made, it increases to 100% when each pattern is subdivided into 5 or 10 subpatterns. However, the former scenario of no pattern subdivision is more jitter resilient than the later ones.

Online, real-time learning in neuromorphic circuits have been implemented through variants of Spike Time Dependent Plasticity (STDP). Current implementations have used either floating-gate devices or memristors to implement such learning synapses together with non-volatile storage. However, these approaches require high voltages (≈ 3-12V) for weight update and entail high energy for learning (≈ 4-30pJ/write). We present a domain wall memory based low-voltage, low-energy STDP synapse that can operate with a power supply as low as 0.8V and update the weight at ≈ 40fJ/write. Device level simulations are performed to prove its feasibility. Its use in associative learning is also demonstrated by using neurons with dendritic branches to classify spike patterns from MNIST dataset.

In this letter, we propose a novel neuro-inspired low-resolution online unsupervised learning rule to train the reservoir or liquid of liquid state machines. The liquid is a sparsely interconnected huge recurrent network of spiking neurons. The proposed learning rule is inspired from structural plasticity and trains the liquid through formating and eliminating synaptic connections. Hence, the learning involves rewiring of the reservoir connections similar to structural plasticity observed in biological neural networks. The network connections can be stored as a connection matrix and updated in memory by using address event representation (AER) protocols, which are generally employed in neuromorphic systems. On investigating the pairwise separation property, we find that trained liquids provide 1.36 0.18 times more interclass separation while retaining similar intraclass separation as compared to random liquids. Moreover, analysis of the linear separation property reveals that trained liquids are 2.05 0.27 times better than random liquids. Furthermore, we show that our liquids are able to retain the generalization ability and generality of random liquids. A memory analysis shows that trained liquids have 83.67 5.79 ms longer fading memory than random liquids, which have shown 92.8 5.03 ms fading memory for a particular type of spike train inputs. We also throw some light on the dynamics of the evolution of recurrent connections within the liquid. Moreover, compared to separation-driven synaptic modification', a recently proposed algorithm for iteratively refining reservoirs, our learning rule provides 9.30%, 15.21%, and 12.52% more liquid separations and 2.8%, 9.1%, and 7.9% better classification accuracies for 4, 8, and 12 class pattern recognition tasks, respectively.

In this brief, a neuron with nonlinear dendrites (NNLDs) and binary synapses that is able to learn temporal features of spike input patterns is considered. Since binary synapses are considered, learning happens through formation and elimination of connections between the inputs and the dendritic branches to modify the structure or morphology of the NNLD. A morphological learning algorithm inspired by the tempotron, i.e., a recently proposed temporal learning algorithm is presented in this brief. Unlike tempotron, the proposed learning rule uses a technique to automatically adapt the NNLD threshold during training. Experimental results indicate that our NNLD with 1-bit synapses can obtain accuracy similar to that of a traditional tempotron with 4-bit synapses in classifying single spike random latency and pairwise synchrony patterns. Hence, the proposed method is better suited for robust hardware implementation in the presence of statistical variations. We also present results of applying this rule to real-life spike classification problems from the field of tactile sensing.

We present the current mode implementation of a spiking neural classifier with lumped square law dendritic nonlinearity. It has been shown earlier that such a system with binary synapses can be trained with structural plasticity algorithms to achieve comparable classification accuracy with less synaptic resources than conventional algorithms. Hence, in our address event based implementation, we save 2-12X memory resources in storing connectivity information. The chip fabricated in 0.35μm CMOS has 8 dendrites per cell and uses two opposing cells per class to cancel common mode inputs. Preliminary results show the chip is functional and dissipates 30nW of static power per neuronal cell and 422pJ/spike.

The linear antenna array design problem is one of the most important in electromagnetism. While designing a linear antenna array, the goal of the designer is to achieve the “minimum average side lobe level” and a “null control” in specific directions. In contrast to the existing methods that attempt to minimize a weighted sum of these two objectives considered here, in this paper our contribution is twofold. First, we have considered these as two distinct objectives which are optimized simultaneously in a multi-objective framework. Second, for directivity purposes, we have introduced another objective called the “maximum side lobe level” in the design formulation. The resulting multi-objective optimization problem is solved by using the recently-proposed decomposition-based Multi-Objective Particle Swarm Optimizer (dMOPSO). Our experimental results indicate that the proposed approach is able to obtain results which are better than those obtained by two other state-of-the-art Multi-Objective Evolutionary Algorithms (MOEAs). Additionally, the individual minima reached by dMOPSO outperform those achieved by two single-objective evolutionary algorithms.

Differential evolution (DE) is one of the most powerful stochastic real parameter optimizers of current interest. In this paper, we propose a new mutation strategy, a fitness-induced parent selection scheme for the binomial crossover of DE, and a simple but effective scheme of adapting two of its most important control parameters with an objective of achieving improved performance. The new mutation operator, which we call DE/current-to-gr_best/1, is a variant of the classical DE/current-to-best/1 scheme. It uses the best of a group (whose size is q% of the population size) of randomly selected solutions from current generation to perturb the parent (target) vector, unlike DE/current-to-best/1 that always picks the best vector of the entire population to perturb the target vector. In our modified framework of recombination, a biased parent selection scheme has been incorporated by letting each mutant undergo the usual binomial crossover with one of the p top-ranked individuals from the current population and not with the target vector with the same index as used in all variants of DE. A DE variant obtained by integrating the proposed mutation, crossover, and parameter adaptation strategies with the classical DE framework (developed in 1995) is compared with two classical and four state-of-the-art adaptive DE variants over 25 standard numerical benchmarks taken from the IEEE Congress on Evolutionary Computation 2005 competition and special session on real parameter optimization. Our comparative study indicates that the proposed schemes improve the performance of DE by a large magnitude such that it becomes capable of enjoying statistical superiority over the state-of-the-art DE variants for a wide variety of test problems. Finally, we experimentally demonstrate that, if one or more of our proposed strategies are integrated with existing powerful DE variants such as jDE and JADE, their performances can also be enhanced.

Orthogonal Frequency Division Multiplexing (OFDM) has emerged as very popular wireless transmission technique in which digital data bits are transmitted at a high speed in a radio environment. But the high peak-to-average power ratio (PAPR) is the major setback for OFDM systems demanding expensive linear amplifiers with wide dynamic range. In this article, we introduce a low-complexity partial transmit sequence (PTS) technique for diminishing the PAPR of OFDM systems. The computational complexity of the exhaustive search technique for PTS increases exponentially with the number of sub-blocks present in an OFDM system. So we propose a modified Differential Evolution (DE) algorithm with novel mutation, crossover as well as parameter adaptation strategies (MDE_pBX) for a sub-optimal PTS for PAPR reduction of OFDM systems. MDE_pBX is utilized to search for the optimum phase weighting factors and extensive simulation studies have been conducted to show that MDE_pBX can achieve lower PAPR as compared to other significant DE and PSO variants like JADE, SaDE and CLPSO.

Time Division Multiple Access (TDMA) plays an important role in MAC (Medium Access Control) for wireless sensor networks providing real-time guarantees and potentially reducing the delay and also it saves power by eliminating collisions. In TDMA based MAC, the sensor are not allowed to radiate signals when they are not engaged. On the other hand, if there are too many switching between active and sleep modes it will also unnecessary waste energy. In this paper, we have presented a multi-objective TDMA scheduling problem that has been demonstrated to prevent the wasting of energy discussed above and also further improve the time performance. A Modified Discrete Differential Evolution (MDDE) algorithm has been proposed to enhance the converging process in the proposed effective optimization framework. Simulation results are given with different network sizes. The results are compared with the Particle Swarm Optimization (PSO) and Genetic Algorithm (GA) and the original Discrete DE algorithm (DDE). The proposed MDDE algorithm has successfully outperformed these three algorithms on the objective specified, which is the total time or energy for data collection.